This article has been accepted for publication in IEEE Transactions on Fuzzy Systems.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

Fuzzy Machine Learning:

A Comprehensive Framework and Systematic Review
Jie Lu, Fellow, IEEE, Guangzhi Ma, Student Member, IEEE, and Guangquan Zhang

Abstract—Machine learning draws its power from various FML stands out as an invaluable ally in the realm of
disciplines, including computer science, cognitive science, and complex, and dynamic (uncertain) environments, presenting
statistics. Although machine learning has achieved great ad- substantial advantages that elevate its efficacy. Unlike tra-
vancements in both theory and practice, its methods have some
limitations when dealing with complex situations and highly ditional machine learning approaches, fuzzy techniques that
uncertain environments. Insufficient data, imprecise observations, generally based on the concept of fuzzy sets [9] and fuzzy
and ambiguous information/relationships can all confound tra- theory [10] excel in capturing and navigating the nuanced
ditional machine learning systems. To address these problems, shades of uncertainty inherent in dynamic scenarios. Their
researchers have integrate machine leaning from different as- inherent ability to model uncertainty empowers it to gracefully
pects, and fuzzy techniques including fuzzy sets, fuzzy systems,
fuzzy logic, fuzzy measures, fuzzy relations, and so on. This adapt to the ever-changing patterns that characterize dynamic
paper presents a systematic review of fuzzy machine learning, environments. In situations where traditional models might
from theory, approach to application, with the overall objective falter or struggle to keep pace, fuzzy techniques emerge as
of providing an overview of recent achievements in the field of robust problem-solvers, providing a more accurate represen-
fuzzy machine learning. To this end, the concepts and frameworks tation of the inherent fuzziness present in real-world data
discussed are divided into five categories: (a) fuzzy classical
machine learning; (b) fuzzy transfer learning; (c) fuzzy data [11]. Furthermore, in the relentless quest for interpretability,
stream learning; (d) fuzzy reinforcement learning; and (e) fuzzy fuzzy machine learning triumphs. Its models not only navigate
recommender systems. The literature presented should provide complexity but also offer clear insights into decision-making
researchers with a solid understanding of the current progress processes. This interpretability proves to be a critical asset
in fuzzy machine learning research and its applications. in dynamic environments where understanding the rationale
Index Terms—Machine learning, fuzzy sets and systems, fuzzy behind model decisions is paramount. Next, we summarize
logic, transfer learning, data stream learning, recommender some main successes of how fuzzy techniques can improve
systems machine learning algorithms:

I. I NTRODUCTION 1) Fuzzy sets [2] can be used to represent vague or am-

biguous concepts and data, such as that commonly found
I N the dynamic realm of technology, machine learning has
profoundly transformed various sectors. It leads innovation
by decoding complex data patterns, driving advancements in
with linguistic variables, noisy or incomplete data, and
interval-valued data. The fuzzy sets enhance the algo-
rithm’s ability to make decisions in uncertain and com-
artificial intelligence, and influencing how we engage with
plex situations, which can be particularly useful in appli-
information and understand the capabilities of computational
cations where real-world conditions can be unpredictable,
systems. However, with most existing machine learning meth-
such as robotics or autonomous vehicles.
ods, accuracy suffers in scenarios characterized by uncertainty,
2) Fuzzy rule-based systems [3] can provide a transpar-
such as the only available observations are imprecise, or where
ent and interpretable prediction frameworks. Fuzzy rule-
the data are noisy or incomplete. Additionally, many real-
based systems use linguistic rules to represent knowledge
world datasets contain uncertain relationships, and conven-
and, so, can be used to generate explanations for the
tional machine learning methods generally find it difficult
decisions made by the system. This can be useful in
to identify or work with these structures. To address these
applications like medical diagnosis.
issues, researchers have used fuzzy techniques to integrate into
3) Fuzzy clustering [4], which is a well-known approach
machine learning called Fuzzy Machine Learning [1] (FML) as
to clustering, can improve machine learning algorithms
a solution, since fuzzy techniques are successful to deal with
by identifying patterns in data that traditional clustering
uncertainties. FML systems fuse machine learning algorithms
methods may not easily identify. Fuzzy clustering not
with fuzzy techniques, such as fuzzy sets [2], fuzzy systems
only allows for overlapping clusters, it can also handle
[3], fuzzy clustering [4], fuzzy relations [5], fuzzy measures
data points that may not belong to any particular cluster
[6], fuzzy matching [7], fuzzy optimization [8], and so on, to
with certainty. This can be useful in applications like
build new models that are more robust to the many and varied
image recognition.
types of uncertainty found in real-world problems.
4) Fuzzy relations [5] can provide a more flexible and
The work presented in this paper was supported by the Australian Research nuanced representation of the relationships between vari-
Council (ARC) under FL190100149 and DP220102635. ables or data points. They can also capture nonlinear
Jie Lu, Guangzhi Ma and Guangquan Zhang are with the Australian Artifi- relationships to enable more accurate and expressive
cial Intelligence Institute, Faulty of Engineering and Information Technology,
University of Technology Sydney, Sydney, NSW, 2007, Australia, e-mail: machine learning models. In addition, fuzzy relations are
[Link]@[Link], {[Link]; [Link]}@[Link]. useful when handling multi-modal data or data assembled

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

No-deep learning-based
from multiple sources because researchers can define Fuzzy classical machine learning
Deep learning-based
fuzzy relations between the different modalities to result
Fuzzy set-based
in a more comprehensive and accurate model.
Fuzzy transfer learning Fuzzy systems-based
In the past decade, there have been over 500, 000 articles in Fuzzy relation-based
high-quality journals and conference proceedings containing
Detection
Comprehensive
the words “fuzzy” and “machine learning”. However, none of framework of
Fuzzy data stream learning Adaptation
fuzzy machine
these articles provide a comprehensive review of the recent learning
Both detection and adaptation
literature on FML. Several previous surveys in the area only
Control systerms
offered valuable insights into certain subfields of FML. For Fuzzy reinforcement learning
Multi-agent
example, Baraldi et al. [12] provides a brief review of fuzzy
Collaborative filtering-based
clustering algorithms for pattern recognition, while Skrjanc et
Cotent-based
al. [13] summarizes models based on evolving fuzzy rules and Fuzzy recommender systems
Knowledge-based
neuro-fuzzy networks for clustering, regression, identification,
Hybrid
and classification problems. Additionally, Zheng et al. [14]
reviewed recent work on fusing deep learning models with
Fig. 1. The main framework of this survey.
fuzzy systems. Moreover, the last decade has witnessed the
emergence of new subfields in FML, such as fuzzy transfer
learning and fuzzy data stream learning. Providing an inves- used to address these issues. The insights provided are
tigative report to outline these new subfields is significant. designed to help researchers understand the context of
For these reasons, a new, more comprehensive, and more up- developments in FML research and its applications.
to-date survey of FML is warranted. This paper primarily 3) It provides a critical discussion of the state-of-the-art
targets researchers interested in employing fuzzy techniques (SOTA) FML models and outlines directions for future
to enhance the performance of machine learning methods, par- research.
ticularly in situations involving complex or uncertain factors. The remainder of this paper is structured as follows. Section
The studies included in this survey were selected following II provides some relevant mathematical concepts to illustrate
three steps: how fuzzy logic can be integrated into machine learning. Sec-
tions III-VII discuss the five categories of FML respectively.
Step 1 Identify and determine an appropriate set of publica-
Finally, Section IX summarizes the material covered and goals
tion databases to search. We searched the well-known
of this review and outlines future work.
databases of Science Direct, ACM Digital Library,
IEEE Xplore and SpringerLink. These provided a com-
prehensive bibliography of research papers on machine II. BASIC C ONCEPTS OF F UZZY M ACHINE L EARNING
learning and FML. In this section, we briefly introduce some relevant mathe-
Step 2 Preliminary screening of articles: The first search was matical concepts to illustrate how fuzzy logic can be integrated
based on keywords. The articles were then selected into transfer learning, data stream learning, reinforcement
for inclusion in the review if they: 1) presented a learning, and recommender systems. These concepts should
new theory, algorithm or methodology in the area of help researchers to better understand the articles introduced in
FML; or 2) reported an application built around a FML the following sections.
algorithm.
Step 3 Filtering the results for presentation: The articles se- A. Fuzzy transfer learning
lected in Step 2 were then divided into five groups to
Transfer learning [15] tries to train a well-performed model
be summarized in separate sections: (a) fuzzy classical
in one domain (target) by leveraging knowledge from another
machine learning; (b) fuzzy transfer learning; (c) fuzzy
domain (source) that has different distribution or learning tasks
data stream learning; (d) fuzzy reinforcement learning;
compared with the previous one. This section introduces two
and (e) fuzzy recommender systems. At this point, we
representative fuzzy transfer learning frameworks: 1) fuzzy
undertook one final screening of the articles, see Fig.
rule-based [16]; 2) fuzzy equivalence-based [17].
1. A study was retained if it demonstrated sufficient:
Fuzzy rule-based transfer learning framework [16]: Let
(1) novelty, i.e., it had been published within the last
S ={S 1 , S 2 , · · · , S N } denotes a set of source domains, where
decade; and (2) impact, i.e., it had been published
S n = {(xSi n , yiSn )|xSi n ∈ X n , yiSn ∈ Y}m i=1 , n ∈ [N ] and
n
in a high-quality journal/conference or having high Sn Sn
(xi , yi ) is the i-th input-output data pairs in the n-th source
citations.
domain. Here, X n ⊂ Rp denotes the feature space of each
The main contributions of this paper are as follows: source domain and Y is a response space (Y = {1, 2, · · · , K}
1) It survey comprehensively summarizes the developments given a classification task, and Y ⊂ R given a regression task).
and achievements in the field of FML. Work in this field T = {xTi |xTi ∈ X T }m t
i=1 is the unlabeled target domain (for
is divided into five main categories for discussion. unsupervised scenario), where X T ⊂ Rp is the feature space
2) The shortcomings of traditional machine learning meth- of the target domain. In homogeneous cases, X 1 , · · · , X N , X T
ods in real-world scenarios are analyzed for each cate- have the same number of features, while they contain different
gory, followed by an explanation of how FML has been number of features in heterogeneous cases.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

We denote R = {R1 , R2 , · · · , RN } as the constructed data instances and the membership degree of each instance
fuzzy rules space of S, where Rn = {r(vlSn , aSl n )}ll=1
n
,n ∈ belonging to each pattern during the process of learning the
[N ] is the n-th rule set of S . Here, the rule r(vl , aSl n ) is
n Sn
parameters for the predictor. Given {µtk } the membership of
represented as: t-th instance belonging to k-th cluster, {Ck } the k-th cluster
centroid, {Xt } the input variables at time step t and {θt } the
if xSi n is Al (xSi n , vlSn ),
parameters for the k-th predictor. Then, the purpose of fuzzy
then yiSn is Pl (xSi n , aSl n ), (1) clustering-based drift learning is shown as follow:
l = 1, 2, · · · , ln . N X
X K

Let RT denote the obtained fuzzy rules of target domain T . min ( µtk Xt θk − yt )2 (3)
µtk ,Ck ,θt
t=1 k=1
Finally, Φ = {Φ1 , Φ2 , · · · , ΦN } is denoted as the conclu-
K X
N K
sion of R (eg., linear combination), where Φn (Rn , µn ), n ∈ X X
+ λ1 µ2tk ∥ Xt − Ck ∥22 +λ2 ∥ θk ∥22
[N ] is the n-th conclusion of Rn . Hence, fuzzy rule-based
k=1 t=1 k=1
transfer learning aims to use the knowledge from D = K
{S, R, Φ} to fit the data in the target domain, i.e., obtain RT s.t.
X
µtk = 1, t ∈ [N ].
and the conclusion of RT . k=1
Fuzzy equivalence-based transfer learning framework
[17]: Different from fuzzy rule-based transfer learning, this where λ1 , λ2 are two pre-assigned parameters. Fuzzy cluster-
framework applies the fuzzy equivalence relations among ing [25], [26] is utilize to optimize µtk , Ck .
features in source and target domains to replace the fuzzy
rules. Let U = {U 1 , U 2 , · · · , U N } denotes the member- C. Fuzzy reinforcement learning
ship function space of the features in S, where U n = Reinforcement learning (RL) [27] is the study of planning
{µS1 n , µS2 n , · · · , µSmnn }, n ∈ [N ] and µSi n , i ∈ [mn ] is the and learning in a scenario where a learner (called an agent)
membership function of xSi n . RM M M
S = {R1 , R2 , · · · , RN }
M
proactively interacts with the environment to achieve a certain
denoted as the fuzzy equivalence relations space on S, where goal. The agent’s aim is to develop the optimal strategy for
RM n , n ∈ [N ] is the fuzzy equivalence relation on S . Here,
n
accumulating rewards. It does this by learning from the feed-
RM n , n ∈ [N ] is a m n × mn matrix (see [17], [18] for details), back it receives. Reinforcement learning has been successfully
(RM Sn Sn Sn Sn applied to a variety of real-world problems, such as robotics
n )ij = RSn (xi , xj ; µi , µj ), i, j ∈ [mn ], (2)
control [28], game playing [29], and autonomous driving [30].
where RSn is a fuzzy equivalence relation operator on S n . In this section, we provide information of how fuzzy logic can
Hence, fuzzy equivalence-based transfer learning framework be integrated into RL.
aims to use the knowledge from D = {S, U, RM S } to fit the First, fuzzy sets can be used to represent uncertainty in state,
data in the target domain. action, or reward spaces in RL. For instance, fuzzy reward
signals [31] represent the uncertainty or imprecision in the
B. Fuzzy data stream learning reward received by an agent. Additionally, fuzzy controllers
Data stream learning [19], [20], also known as stream [32] that use fuzzy logic to map inputs to control actions can
mining, refers to a set of techniques and algorithms designed be integrated into RL systems to handle uncertain or qualita-
to handle and analyze data that arrives continuously over tive control decisions. Next, we give a general mathematical
time in a streaming fashion. However, in real-world scenarios, expression for fuzzy controller. Let X1 , · · · , Xn be the input
the statistical properties of the data may change over time, variables to the fuzzy controller, and Y be the output variable
making models and algorithms that were previously accurate representing the control action. The fuzzy sets associated with
less effective over time. This phenomenon is known as concept each variable are denoted as A1 , · · · , An for inputs and B for
drift [21]–[23]. A formal definition of concept drift follows. the output. Let µAi (xi ) represent the membership function
Definition 1 (Concept drift [23]): Consider a time period for the fuzzy set Ai of input Xi , and µB (y) represent the
[0, t] and a set of samples, denoted as S0,t = {d0 , · · · , dt }, membership function for the fuzzy set B for output Y . Then,
where di = (Xi , yi ) is one observation (or one data in- generic fuzzy rules that define the mapping from inputs to
stance). Xi is the feature vector, yi is the label, and S0,t outputs can be expressed as:
follows a certain distribution F0,t (X, y). Concept drift occurs if Xi is A1j and · · · and Xn is Anj , then Y is Bj . (4)
at timestamp t + 1, if F0,t (X, y) ̸= Ft+1,∞ (X, y), denoted as
∃t : Pt (X, y) ̸= Pt+1 (X, y). After applying the fuzzy rules, defuzzification is performed to
Hence, when a concept drift occurs at t+1, we aim to adapt obtain a crisp output value.
the predictor Ht = arg minh∈H ℓ(h, X, y|(X, y) ∈ Pt (X, y)) Further, a fuzzy inference system can be used to make
to fit the new distribution Pt+1 (X, y). Next, we briefly in- decisions in RL [33], such as determining the next action
troduce a fuzzy clustering-based drift learning structure [24] to take based on fuzzy input signals representing uncertain
to show how fuzzy logic can be integrated into data stream states or rewards. For example, fuzzy Q-learning [34] extends
learning. Q-learning by incorporating fuzzy logic to handle uncertain
In fuzzy clustering-based drift learning [24], fuzzy cluster- and imprecise state-action pairs. Fuzzy rules and membership
ing is applied to learn how many patterns exist in the observed functions are applied to update the Q-values.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

D. Fuzzy recommender system 1) Clustering: Fuzzy clustering has been widely researched
A recommender system [35], [36] is a type of information over the last 40 years, and several survey papers have already
filtering system that analyzes user preferences or behavior to been published summarizing prior work in this field [39], [40].
provide suggestions personalized to that particular user. In this First, we summarize the main ascendancies of applying fuzzy
section, we provide information of how fuzzy logic can be techniques in clustering.
integrated into recommender systems. • Soft Assignment of Data Points: Traditional clustering
Let V = {v1 , v2 , · · · , vM } denotes the item set and U = algorithms assign each data point to a single cluster,
{u1 , u2 , · · · , uN } denotes the user set. In data preprocess, resulting in a hard assignment. However, in many cases,
fuzzy set or linguistic variable can be used to represent some of the data points may have ambiguous relation-
item/user terms and user-item rating matrix R, (R)ij = ships with the clusters or their memberships may overlap
rui ,vj , i ∈ [N ], j ∈ [M ] (rui ,vj is a rating of a user for a item). into multiple clusters. Fuzzy clustering allows for soft
Fuzzy set can help dealing with some types of uncertainty assignment, where a data point’s membership in a cluster
in the description of item features. For example, Yager [37] is not simply binary, but rather it is measured in degrees
denotes a set of primitive assertions to describe items, denoted and can apply to multiple clusters.
as A = {A1 , · · · , An }. For one item v, we can view the item • Flexibility in Cluster Shape: Unlike traditional hard
v as a fuzzy subset over the space A. If one item v satisfies clustering algorithms, such as K-means, which assume
assertion Ai , the assertion has validity equal to one otherwise spherical clusters of equal size, fuzzy clustering allows
zero. The membership degree on Ai in v is v(Ai ). Then, an for more flexible and irregular cluster shapes. Fuzzy logic
item in a recommender system can be represented as a fuzzy allows researchers and analysts to model overlapping
set over an assertion set. In addition, linguistic variable is clusters, clusters of varying sizes and densities, and
widely used to generate the user-item linguistic term-based clusters with complex boundaries. Thus, fuzzy clustering
rating matrix R. is highly suitable for datasets with complex structures.
In fuzzy user preference/profile generation process, fuzzy • Handling Outliers and Noise: Applying fuzzy logic
rule-based system, such as Takagi–sugeno–kang fuzzy system makes clustering more robust to outliers and noisy data
(TSK-FS), is usually applied to model the uncertainty and than traditional clustering methods. With fuzzy logic, a
imprecision inherent in users’ preferences. Finaly, in order data point can have a low membership degree to a cluster,
to obtain the final predicted ratings r̂ui ,vj for unrated items, which effectively reduces the influence of outliers or
fuzzy similarity is widely used for calculating the similarity noisy data points on the overall clustering results.
between items and users. For instance, S(vi , vj ) [38] is a fuzzy • Interpretability and Granularity: The fuzzy member-
similarity to measure the similarity between item vi and vj , ship degrees assigned to data points offer a quantitative
measure of their association with each cluster. This allows
P R1
0
f ([ru,vi ]α , [ru,vj ]α )dα for a more nuanced understanding of the data and pro-
S(vi , vj ) = r P
u∈Uij
r P , vides insights into the degree of similarity or dissimilarity
R1 R1
( 0
g([ru,vi ]α )dα)2 ( 0
h([ru,vj ]α )dα)2 between data points and clusters. Fuzzy logic also allows
u∈Uij u∈Uij for the representation of gradual transitions, providing a
(5) more detailed and fine-grained view of the clustering.
where Uij represents the set of users that both rated items One of the most powerful and well-known algorithms in fuzzy
vi and vj . [ru,vi ]α represent the α-cut of ru,vi (linguistic clustering analysis is Fuzzy C-means (FCM), developed by
variable), and f, g, h are predefined functions. Dunn in 1973 [25] and further developed by Bezdek in 1984
[41]. In the intervening years, FCM has been widely used and
III. FUZZY CLASSICAL MACHINE LEARNING revised many times to deal with different types of problems
Classical machine learning algorithms, such as decision [42]–[46]. Among the most recent of these achievements,
trees, support vector machines, and neural networks, have been Ding et al. [42] proposed a novel kernel-based FCM clus-
responsible for remarkable achievements both theoretically tering algorithm that uses genetic algorithm optimization to
and from a practical point of view. Numerous articles involve improve clustering performance. To enhance the robustness of
combining fuzzy techniques with classical machine learning image segmentation, Gao et al. [46] presented a new robust
algorithms to overcome different types of problems with FCM clustering method that combines an elastic FCM with
uncertainty, such as incomplete information and imprecise ob- a smoothing method. This elastic FCM provides a sparser
servations. In this section, we summarize these works, dividing description for reliable points and a fuzzier description of the
the techniques into two categories: 1) Non-deep learning-based marginal points of clusters. Lei et al. [43] designed a more
method, 2) Deep learning-based method. efficient and more robust FCM algorithm for fast and reliable
image segmentation. Their variant is based on morphological
reconstruction and membership filtering. Subsequently, Lei et
A. Non-deep learning-based method al. [47] built a fuzzy clustering framework around the above
The non-deep learning-based methods can be further divided implementation for image segmentation.
into three main types: clustering, regression, and classification. In research departing from FCM, Jiao et al. [48] devel-
Each is discussed in turn next. oped a fuzzy clustering algorithm that relies on unsupervised

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

fuzzy decision trees to improve model interpretability. To

Input data Fuzzification Defuzzification Output data
cluster multiple nominal data streams, Sangma et al. [49] DNN

proposed a fuzzy hierarchical clustering method that involves

the clustering-by-variable approach. The method calculates
Learning algorithm
the fuzzy affinity of data streams to different clusters using
normalized cosine similarity and handles concept evolution
Input data Fuzzy rule-based
Output data
by updating the hierarchical clustering structure. DNN inference system

2) Regression: Fuzzy regression models [50] perform a

type of regression analysis that incorporates both possibility Learning algorithm
theory and fuzzy set theory [51]. They are particularly useful
when precise data is lacking or when the relationships between Fig. 2. Two different FNN structures.
input and output variables are complex and difficult to model
using classical regression methods. In addition, fuzzy regres- softmin model based on an enhanced TSK-FS to classify high-
sion models are good at expressing nonlinear relationships and dimensional datasets. An adaptive softmin function overcomes
dealing with noisy data. Fuzzy logic handles the nonlinear the drawbacks of “numeric underflows” and “fake minimums”
relationships between variables. Noisy or incomplete data is that frequently arise in existing fuzzy systems. However, al-
handled by allowing for partial memberships and fuzzy sets. though the enhanced TSK-FS maintains adequate rules, it does
Thus, by assigning lower membership values to outliers or not grow the number of rules exponentially with features. Ma
inconsistent data points, fuzzy regression models provide a et al. [11], [84] put forward a novel framework for addressing
mechanism for mitigating the impact of this type of uncer- multi-class classification problems with imprecise observations
tainty. Another technique for improving the performance of that provides a theoretical analysis of the problem based on
regression models has been to integrate fuzzy techniques with fuzzy Rademacher complexity. The imprecise observations can
classical machine learning techniques, such as SVM [52]–[54] be either fuzzy-valued or interval-valued, and the framework,
and neural networks [55]–[58]. Other solutions combine inter- which combines classical machine learning techniques like
val regression analysis [59] with machine learning methods neural networks and SVM with a fuzzy membership-based
[60]–[63]. defuzzification method, extracts crisp-valued information from
The latest developments in fuzzy regression analysis include these fuzzy- or interval-valued features.
He et al. [64], [65], who developed a fuzzy nonlinear regres-
sion model using a random weight network that takes triangu-
B. Deep learning-based method
lar fuzzy numbers as its inputs and outputs the same. Baser and
Demirhan [66] proposed a new method that combines fuzzy Fuzzy neural networks (FNNs), also known as neuro-fuzzy
regression models with SVM to estimate the yearly mean and networks (NFN), are a type of hybrid neural network that
daily values of horizontal global solar radiation. By applying combine fuzzy techniques with neural networks to improve
fuzzy regression functions, their method is robust to outlier the efficiency and interpretability of machine learning models.
observations and problems with over-fitting. Chachi [67] de- A standard FNN structure is illustrated in Fig. 2. The fuzzy
signed a robust fuzzy regression modeling technique based logic component of FNNs allow them to handle imprecise or
on weighted least squares fuzzy regression to handle crisp incomplete data and make decisions based on uncertain inputs.
input-fuzzy output data. Choi et al. [68] addressed issues with Additionally, using an FNN capitalizes on the many significant
multicollinearity in fuzzy regression models by incorporating advances achieved through deep learning in fields such as
ridge regression. Naderkhani et al. [69] proposed an adaptive computer vision, natural language processing, and robotics.
neuro-fuzzy inference system for analyzing and predicting Many researchers have fused deep learning methods with
nonparametric fuzzy regression functions with crisp-valued fuzzy techniques to address different types of problems with
inputs and symmetric trapezoidal fuzzy outputs. Xia et al. [70] uncertainty. The most commonly used deep learning mod-
developed a novel regression model built around a Takagi- els include deep belief networks (DBNs) [85], convolutional
Sugeno fuzzy regression tree to address complex industrial neural networks (CNNs) [86], and recurrent neural network
modeling problems, while Zhang et al. [71] introduced an in- (RNNs) [87]. In this section, we summarize the SOTA achieve-
terpretable model based on graph community neural networks ments in FNNs from 2020 to 2023. Earlier research successes
and time-series fuzzy decision trees for predicting the delays can be found in prior surveys like [14], [88], [89].
experienced by a high-speed train. Chen et al. [90] devised a fuzzy deep neural network with
3) Classification: Numerous studies have combined fuzzy a sparse autoencoder as a way to try and predict human
techniques with classical machine learning algorithms to ad- intentions. The model is based on human emotions and identi-
dress classification problems. The techniques used include fication information. Lu et al. [91] constructed a novel hashing
fuzzy decision trees [72]–[76], neuro-fuzzy classification [77]– method that integrates deep neural networks and fuzzy logic to
[79], and SVR-based fuzzy classification [80], [81]. Rabcan et measure the similarity between pairwise images. Zadeh [51]
al. [82], for example, recently introduced a new approach to introduced the concept of a type-2 fuzzy set as far back as
signal classification that includes a fuzzification procedure in 1975. These sets, whose membership level themselves are
the transformation process and fuzzy decision trees to perform type-1 fuzzy sets, can be used when there is uncertainty about
the classifications. Xue et al. [83] proposed an adaptive the membership function itself – for example, if one does not

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

TABLE I
S UMMARY OF THE SOTA DEEP LEARNING - BASED FUZZY NEURAL NETWORKS transferred. These knowledge categories include: instances
ACHIEVEMENTS . [159], feature representations [160], model parameters [161],
and relational knowledge [162]. Alternatively, in terms of
Year Fuzzy technique Deep learning algorithm No.
the problem settings tackled, studies can be grouped into
CNN Others(eg., RNN) four categories: multi-task learning [163], domain adaptation
2020 Fuzzy clustering [94] [107] [108] [109] - 4 [164], [165], cross-domain adaptation [166], and heteroge-
Type-1 fuzzy systems [94] [108] [110] [91] [111] [101] [112] [113] 8 neous learning [167]. We have divided our summary of recent
Type-2 fuzzy systems [107] [114] [115] - 3 works (2015 to 2023) into three areas based on the fuzzy
Others [116] [116] [117] [118] [119] 4 technique used. These are fuzzy sets, fuzzy systems, and fuzzy
2021Type-1 fuzzy systems [120] [121] [122] [123] [124] [125] 6 relations. Table II summarizes recent achievements in the field
Type-2 fuzzy systems - [126] [127] [128] 3 of fuzzy transfer learning.
Fuzzy entropy [129] [129] 1
2022 Fuzzy clustering - [130] [131] 2
A. Transfer learning based on fuzzy sets
Type-1 fuzzy systems [132] [133] [134] [135] [132] [130] [138] [139] 13
[136] [137] [140] [141] [142] [143] Behbood et al. [168] proposed an innovative fuzzy-based
Others [144] [145] - 2 transfer learning framework to predict long-term bank failures.
2023 Fuzzy clustering [146] [147] 2 The framework relies on fuzzy sets, as well as similarity
Type-1 fuzzy systems [148] [149] [146] [150] [147] [153] [154] [155] 11 and dissimilarity, to modify the labels of target instances
[151] [152] [156] [157]
predicted by an FNN classifier. Wu et al. [169] developed
know the shape of the function or some of its parameters. The OwARR, a new algorithm that combines fuzzy sets with
superior performance of type-2 fuzzy sets has seen them used domain adaptation. The aim is to reduce the amount of object-
in a range of machine learning tasks. For instance, to perform specific calibration data so as to solve the important regression
complex stock time series tasks, Cao et al. [92], [93] designed problem of estimating online drowsiness in drivers from EEG
two multi-objective evolution models. Both combine interval signals in brain-computer interfaces. Gargees et al. [170]
type-2 fuzzy sets with rough FNNs. proposed a transfer learning method for the possibilistic c-
Several fuzzy-based ensemble models have also been de- means clustering problem with insufficient data, overcoming
veloped to address problems like load forecasting [94], image a crucial problem for clustering tasks where the source and
classification [95]–[98], and image fusion [99], [100]. For target domains have a different number of clusters. Based
example, Khatter et al. [101] combined an RNN with fuzzy on the idea of fuzzy sets, the proposed algorithm employs
techniques and a web blog searching method to enhance historical cluster centers of the data in the source domain as a
classification performance, while Concepcion et al. [102] reference to guide the clustering of data in the target domain.
presented a theoretical analysis of why fuzzy-rough cognitive In terms of applying type-2 fuzzy sets to transfer learning
networks delivered better performance than the state-of-the-art models, Sun et al. [171] proposed a new transfer learning
classifiers. Long short-term memory models (LSTMs) have model to address the uncertainty caused by conflicting im-
also been combined with fuzzy techniques. Some example plications in text sequence recognition. The proposed model
works include [103]–[106]. Table I provides a summary of uses FCM to transform the correspondences among words
the SOTA literature related to deep learning with FNNs. into information granules. By integrating type-2 fuzzy sets
In summary, combining fuzzy techniques with classical into a hidden Markov model, this granular information can
machine learning algorithms is not only useful for solving be used for sequence recognition. To reliably estimate GDP
uncertainty problems, like imprecise or noisy data, it can also from only CO2 emission data, Shukla et al. [172] proposed a
improve the interpretability and robustness of the algorithms. new approach to a kernel extreme learning machine (KELM)
Fuzzy sets are good at handling ambiguity and uncertainty that combines transfer learning with interval type-2 fuzzy sets.
and typically provide a more realistic representation of the Interval type-2 fuzzy sets are used to improve the efficiency
inherent fuzziness and uncertainty present. Additionally, fuzzy of the knowledge transfer. To consider the uncertainty in input
logic tends to improve interpretability. These logics often rely datasets, Kumar et al. [173] presented a novel transfer learning
on rule-based systems, where the rules express relationships approach that incorporates type-1 and interval type-2 fuzzy
between the input variables and the output decisions. The rules sets into a KELM framework. The aim is to predict GDP
can either be derived from expert knowledge, or they can be based on uncertain carbon emissions data.
learned from the data. In general, fuzzy sets have been widely applied to address
uncertainty in data in transfer learning scenarios, and, experi-
IV. F UZZY T RANSFER L EARNING mentally, they have been shown to improve both the efficiency
Notably, most current transfer learning [158]methods have and accuracy of knowledge transfer in comparison to non-
limitations when handling real-world situations with uncer- fuzzy methods.
tainty, such as when only a few labeled instances are available.
To overcome these problems, many researchers have turned to
fuzzy sets and fuzzy logic. B. Transfer learning based on fuzzy systems
Existing studies on transfer learning can be divided into Most existing transfer learning methods have a number
categories based on the type of knowledge that is being of drawbacks. For instance, the performance of model-based

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

transfer learning algorithms is heavily dependent on the se- from the source domain to construct fuzzy rules and then mod-
lected classifier. Additionally, feature-based transfer learning ifies these rules using a nonlinear continuous function based
methods can negatively impact the discriminant information on sigmoid functions to estimate values in the target domain.
and geometric properties of instances from both the source To address any significant difference in the label distribution
and target domains. Further, the lack of interpretability and between the source and target domains, Zuo et al. [187] devel-
an inability to handle uncertainty are two significant flaws. To oped some fuzzy system-based domain adaptation models for
address these issues, researchers have turned to fuzzy rule- classification tasks. In [188], they applied granular computing
based systems to improve interpretability and handle uncer- techniques to transfer learning and proposed a comprehensive
tainty. Notably, TSK-FS [3] has received significant attention domain adaptation framework based on a Takagi-Sugeno fuzzy
in this regard. model to handle three different regression scenarios: one where
Shell et al. [174] proposed FuzzyTL, a novel structure that the source and target domains share different conditions, one
combines transfer learning with a fuzzy rule-based system. where they share different conclusions, and one where both
This structure is designed to bridge the knowledge gap be- apply. Moreover, they identified two issues in fuzzy transfer
tween contexts that lack prior direct contextual knowledge. learning that had not yet been resolved: how to choose an
Meher et al. [175] developed an interpretable domain adapta- appropriate source domain and how to efficiently select labeled
tion method, named the rule-based fuzzy ELM classification data for the target domain when the target data structure is un-
model, that uses a fuzzy inference system to design an balanced. The solutions, which involve an innovative method
ELM architecture for remote sensing image classification. The again based on a Takagi-Sugeno fuzzy model [189], combine
model uses the maximum fuzzy membership grade of features, an infinite Gaussian mixture model with active learning to
which is characterized by class-belonging fuzzification, to improve the performance and generalizability of the initial
construct the fuzzy rules and two rule extraction matrices. model. Li et al. [190] designed a new transfer learning model
Moreover, Deng et al. [176], [177] proposed two novel transfer for multi-source domain adaptation that relies on a fuzzy-rule
learning approaches for regression tasks using the Mamdani- based deep neural network. To address the more challenging
Larsen fuzzy system and TSK-FS coupled with a new fuzzy problem in multi-source domain adaptation where no source
logic algorithm and its objective functions. However, they data is available, Li et al. [191] proposed a new model based
noticed that the antecedent parameters of the TSK-FS model on a deep neural network with fuzzy rules.
constructed in the target domain were directly inherited from Importantly, all the domain adaptation studies mentioned
the source domain, which meant that they could not leverage so far only work when both domains have identical feature
enough knowledge from the source domain. To address this spaces and the same number of fuzzy rules, i.e., they are all
problem, Deng et al. [178] proposed a new transfer learning methods of homogeneous domain adaptation. Zuo et al. [192],
method that contains two knowledge-leveraging strategies to however, devised a novel approach to heterogeneous scenarios
better learn the antecedent and consequent parameters in the based on a Takagi-Sugeno fuzzy model. In this framework,
TSK-FS model. First, they applied an FCM-based clustering fuzzy rules are constructed in the source domain and then
transfer technique to the antecedent parameters, which means transferred to the target domain using canonical correlation
that the antecedent parameters can be learned from both analysis so as to minimize the discrepancy between the feature
the source and target domains. Second, they introduced an spaces of the two domains. This was the first article to solve
enhanced knowledge-leverage mechanism to learn the conse- heterogeneous domain adaptation problems using a fuzzy rule-
quent parameters. Another knowledge-leverage term is then based system. Subsequently, Lu et al. [16] addressed the more
introduced to make more effective use of the knowledge in challenging scenario of when the only available instances to
the source domain. Further, they applied and modified these build the model span multiple source domains. They proposed
methods so that they could be used for analysis in scenarios two novel transfer learning methods for regression tasks based
with insufficient data, such as recognizing EEG signals [179]– on a Takagi-Sugeno fuzzy model – one for when the feature
[184] or with situations involving multiple-source domains spaces are homogeneous and one for when the spaces are
[183]. The aim of transfer representation learning is to learn a heterogeneous. In the former, knowledge from multiple source
shared space that matches the distributions of instances from domains is merged in the form of fuzzy rules, while, in
both domains. However, transfer representation learning based the latter, knowledge is merged in the form of both data
on kernels suffers from some shortcomings, such as a lack of and fuzzy rules. Che et al.’s [193] fuzzy transfer learning
interpretability and difficulties with selecting a kernel function. method addresses multi-output regression problems in both
To overcome these issues, Xu et al. [185] proposed a new homogeneous and heterogeneous scenarios. Their approach
transfer representation learning method that uses TSK-FS in- applies fuzzy rules to accurately capture the commonalities
stead of kernel functions to realize nonlinear transformations. and characteristics of multiple numerical output variables.
In this approach, instances from both domains are transformed In summary, most of the above methods share a common
into a fuzzy feature space to minimize the differences between model construction framework: they begin by constructing a
the distributions. Meanwhile, any discriminant information fuzzy rule-based model on the source data (e.g., a TSK-FS)
or geometric properties are preserved using latent Dirichlet and subsequently modify the existing model (fuzzy rules) to
allocation and principal component analysis. establish a new fuzzy model for the target domain. Fuzzy rule-
Notably, Zuo et al. [186] devised a new way of constructing based systems provide a linguistic representation of knowl-
a TSK-FS model for regression tasks. This model uses data edge, enabling generalization and adaptation, while also mak-

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

TABLE II TABLE III

S UMMARY OF THE SOTA PAPERS IN FUZZY TRANSFER LEARNING . S UMMARY OF THE SOTA ACHIEVEMENTS IN FUZZY DATA STREAM LEARNING .

Fuzzy techniques Type No. Fuzzy techniques Type No.

Regression Classification Detection Adaptation Detection and adaptation Others
Fuzzy sets [169] [172] [173] [168] [171] [197] 6 Fuzzy clustering [201] [202] [203] [204] - - 6
[24] [205]
Fuzzy systems [174] [176] [177] [186] [188] [175] [178] [179] [180] [181] 23
[189] [192] [16] [193] [182] [183] [184] [185] [187] Fuzzy set theory [206] [207] - - - 2
[190] [191] [198] [199]
Fuzzy systems - [208] [209] [212] [213] [214] [215] [216] 12
Fuzzy relations - [18] [196] [17] [200] 4 [210] [211] [217] [218] [219]
Fuzzy time series - [220] [221] - - 2
Others [222] [223] [224] [225] [226] [227] 6
ing the model more robust to domain shift. Their power to
transfer relevant knowledge also helps to improve a model’s researchers tend to apply fuzzy relations to improve transfer
interpretability. All these characteristics make fuzzy rule-based efficiency in heterogeneous situations.
systems well-suited to transfer learning tasks – particularly,
the more challenging tasks, such as heterogeneous domain V. F UZZY DATA S TREAM L EARNING
adaptation and source-free domain adaptation. Learning from data streams [19], [20] involves developing
algorithms and techniques to adaptively and incrementally
process and learn from continuously arriving data. Unlike
C. Transfer learning based on fuzzy relations
traditional machine learning scenarios where a static dataset is
Most studies mentioned so far focus on supervised or semi- available for offline training, data stream learning deals with
supervised transfer learning in homogeneous scenarios, where dynamic, evolving data streams that may not be stored entirely.
both the source and target domains have labeled instances However, data streams often exhibit concept drift, which refers
and only their data distributions are different. However, it is to changes in the statistical properties of the data. Detecting
not uncommon in the real-world for there to be no available and adapting to concept drift are two important challenges in
labeled instances in the target domain. Further, the feature data stream learning. One approach is to continuously monitor
spaces of the source and target domains will usually be differ- the data and update models or retrain them periodically
ent. This scenario, which is characterized by a high degree of to account for changes. Another approach is to use online
uncertainty, is commonly referred to as heterogeneous unsu- learning techniques that can adapt to changes in the data
pervised domain adaptation (HeUDA). Recently, researchers stream in real-time. While concept drift often come with some
have developed n-dimensional fuzzy geometry theory [194] uncertainty problems – for example, making predictions from
and fuzzy equivalence relations [195] to analyze and handle data streams with mixed drift problems and detecting drift in
such problems with uncertainty. data streams with missing values – researchers are considering
Liu et al.’s [18] solution to HeUDA problems, called F- the application of fuzzy techniques to address these challenges.
HeUDA, is to use fuzzy geometry to measure the similarity of The aim of concept drift detection is to identify when
features between the source and target domains. Shared fuzzy concept drift has occurred so that appropriate measures can
equivalence relations are then introduced, which means both be taken to update or retrain the models in question. Several
domains will share the same number of clustering categories. research teams have turned to FCM-based methods to detect
Hence, knowledge can be transferred from a heterogeneous concept drift [201], [202]. These two methods derive fuzzy
source domain to a target domain with only unlabeled data. membership functions from the data stream and use the
Using these techniques, F-HeUDA outperformed the SOTA membership results to mine concept drift patterns. Zhang
models on four real datasets, and performed especially well et al. [206] designed a new drift detection model based on
when the target domain had very few instances. Moreover, Liu fuzzy set theory to address drift problems associated with user
et al. [17], [196] focused on a more realistic problem called the interests for recommender systems, while Dong et al. [207]
multi-source HeUDA problem. Solving this problem involves developed a data distribution-based drift detection method for
transferring knowledge from several different source domains business intelligence and data-driven decision support systems
that have labeled data but heterogeneous dimensions and one that incorporates fuzzy set theory. In both these methods,
target domain with unlabeled data. Their approach, called a fuzzy set theory is used to handle the challenging issue of
shared fuzzy equivalence relations neural network, improves where an item’s features and its related information is usually
upon previous work in shared fuzzy equivalence relations to incomplete and imprecise. Along these lines, Liu et al. [222]
extract the shared fuzzy information contained in multiple proposed a robust drift detection algorithm that can handle
heterogeneous domains. missing values. This algorithm comprises a masked distance
In summary, because there is a high degree of uncertainty learning algorithm to reduce the cumulative errors caused by
when transferring knowledge from a heterogeneous source missing values and a fuzzy-weighted frequency method to
domain to a target domain with only unlabeled data, non-fuzzy identify discrepancies in the data distribution.
models will not usually perform well. Fuzzy relations offer a Concept drift adaptation refers to the process of updating or
flexible, interpretable, and adaptable framework for represent- modifying a machine learning model in response to concept
ing and transferring knowledge between such domains. Hence, drift so that it remains accurate and effective over time. Over

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

TABLE IV
the past five years, many new adaptation models that fuse S UMMARY OF THE SOTA ACHIEVEMENTS IN FUZZY REINFORCEMENT LEARNING .
fuzzy techniques with machine learning algorithms have been
built to deal with the phenomenon of concept drift. The applied Fuzzy techniques Type No.
fuzzy techniques include fuzzy clustering algorithms [24], Control systems Multi-agent RL Others
[203]–[205], fuzzy rule-based systems [208]–[211], and fuzzy Fuzzy systems [231] [232] [233] [234] [235] [236] [237] [238] 8
time series [220], [221]. Song et al. proposed a series of kernel Others [239] - [240] 2
FCM-based adaptive models to handle data stream regression
problems with concept drift [24], [203], [204]. In [203], [204], only can these techniques help to improve the performance
kernel FCM is used to determine the most relevant learning set, of streaming data classification and regression in uncertain
while, in [24], kernel FCM is used to measure the degree to environments, they can also be applied to handle the phe-
which upcoming examples belong to different patterns. These nomenon of concept drift. Table III summarizes these recent
fuzzy membership values are then embedded in the learning achievements in the field of fuzzy data stream learning.
process to handle mixed drift data streams. In summary, fuzzy techniques are applied in data stream
In terms of applying fuzzy rule-based systems, Garcia learning, especially to handle concept drift scenarios, owing to
et al. [210] developed a modified evolving granular fuzzy- their capacity to handle uncertainty, adapt to changing patterns,
rule-based model that incorporates an incremental learning and provide interpretable models. These features make fuzzy
algorithm to simultaneously impute missing data as well as techniques valuable for detecting, understanding, and adapting
adapt the model’s parameters and structure over time. Garcı́a- to concept drift, leading to better performance than non-fuzzy
Vico et al. [211] proposed an evolutionary fuzzy system to methods.
extract knowledge from data streams as a way to adapt to
concept drift. Both these methods use type-1 fuzzy systems; VI. F UZZY R EINFORCEMENT L EARNING
however, by contrast, Pratama et al. [208] proposed an evolv- Reinforcement learning (RL) [27] represents a powerful
ing type-2 recurrent FNN to simultaneously address three paradigm in machine learning, where agents learn to make
challenges: data uncertainty, temporal behavior, and system decisions through interaction with an environment, guided by
absence. Fuzzy time series (FTS) [228] is a mathematical a system of rewards or penalties. However, the traditional
framework that combines fuzzy logic and time series analysis RL framework is not without its challenges, especially in
to model and forecast uncertain and imprecise data over time. scenarios where the training process is inherently slow due to
FTS is particularly useful in situations where the data has complex and uncertain environments [229] or sparse reward
missing values, outliers, or noise, and where traditional time signals [230]. Fuzzy RL emerges as a promising approach to
series models may not perform well. de Lima e Silva et al. address these limitations, leveraging fuzzy logic to enhance
[220] introduced a non-Stationary FTS, while Severiano et training efficiency and overcome the hurdles associated with
al. [221] introduced an evolving forecasting model based on slow reinforcement processes.
FTS to deal with concept drift. Moreover, Liu et al. [223] One of the primary advantages of fuzzy RL lies in its
proposed a new concept drift adaptation method based on adaptability to dynamic (uncertain) environments. In tradi-
a fuzzy windowing approach. Unlike traditional windowing tional RL, slow training processes can be exacerbated by the
methods, this approach employs sliding windows with an challenges posed by dynamic scenarios where the optimal
overlapping period to enable precise identification of the strategy may change rapidly. Fuzzy logic allows the system
data instances that belong to different concepts. Focusing on to gracefully adapt to these changes, incorporating fuzzy
multiple relevant data stream regression with concept drift, rules that capture the gradual transitions and uncertainties
Song et al. [224] developed a new adaptation model based on in the environment. Additionally, in many RL applications,
fuzzy drift variance, where the variance is designed to measure the scarcity of meaningful rewards can impede the learning
the correlated drift patterns among streams. process, leading to slow convergence or even stagnation. Fuzzy
Additionally, several works simultaneously address concept RL introduces the concept of fuzzy rewards [31], enabling the
drift detection and adaptation [212], [213], [225], [226]. For system to consider partial or intermediate successes that may
example, Dong et al. [225] introduced an adaptive ensemble not be fully captured by binary reward signals. This approach
algorithm based on fuzzy instance weighting to handle data helps mitigate the challenge of sparse rewards by providing a
streams involving concept drift. Yu et al. [213] presented an more nuanced and continuous feedback mechanism, allowing
evolving fuzzy-neuro system for streaming data regression that the agent to learn from a broader spectrum of experiences.
employs an online topology learning algorithm to self-organize Fuzzy Sarsa Learning (FSL) [31] is a critic-only fuzzy RL
each layer of the proposed system. To effectively detect drift algorithm that combines the Sarsa algorithm with fuzzy logic.
and adapt the learned model, Zhang et al. [226] proposed a In traditional Sarsa learning, an agent learns to take actions
novel approach that combines a dynamic intuitionistic fuzzy through trial and error that maximize a reward signal in a
cognitive map scheme and a concept drift detection algorithm. given environment. In FSL, the state and action spaces are
More recently, researchers have used fuzzy techniques to represented as fuzzy sets, which allows for a more grad-
address data stream classification and regression problems. ual transition between states and actions, rather than strict
These techniques include evolving fuzzy systems [214]–[216], boundaries. The algorithm updates the Q-value function based
neuro-fuzzy systems [217], granular fuzzy rule-based systems on the fuzzy membership functions of the current state and
[218], [219], and the fuzzy time-matching method [227]. Not action, as well as the fuzzy reward function. The use of fuzzy

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

10

logic allows FSL to handle more complex environments with Data Preprocess Profile Generation Modeling and learning
Recommendation
Generation
uncertain or imprecise information, while still maintaining
fuzzy item user fuzzy similarity Collaborative filtering-
item linguistic terms
a high level of performance. As an example, Fathinezhad representation based

et al. [237] proposed a novel method of robot navigation user neighbours Content-based
filtering
user linguistic terms fuzzy user preference
that combines supervised learning and FSL. Their method
item fuzzy similarty Knowledge-based
applies a zero order Takagi–Sugeno fuzzy controller with some
user-item rating
candidate actions for each rule as the main module. Also, matrix
fuzzy ratings
item neighbours Hybrid

Hein et al. [238] developed a new particle swarm approach

to RL based on fuzzy controllers. This approach builds fuzzy
RL policies by training parameters on world models that Fig. 3. A fuzzy recommender system framework.
simulate real system dynamics. Furthermore, Shi et al. [240]
traffic signal control environments demonstrate the proposed
developed an adaptive fuzzy comprehensive evaluation method
method’s superior performance. Table IV summarizes these
that integrates a fuzzy analytical hierarchy process, a Bayesian
recent achievements in the field of fuzzy RL.
network, and RL. The authors successfully applied this method
In general, by combining RL and fuzzy logic, fuzzy re-
to a robot soccer system, which is a typical complex time-
inforcement learning can handle the complexity of real-world
sequence decision-making system.
environments that involve uncertain information and imprecise
In the field of control systems, Zhang et al. [231] designed
data, making it a promising technique for solving problems in
a new fault-tolerant control algorithm by combining RL with
fields such as robotics, control systems, and game theory.
a fuzzy augmented model for partially unknown systems with
actuator faults. The fuzzy augmented model was inspired
by the well-known Takagi-Sugeno fuzzy model. With this VII. F UZZY R ECOMMENDER S YSTEMS
algorithm, less information needs to be transmitted, which In real-world recommender systems, descriptions of user
reduces computational loads during the learning process, even preferences and item features, item values, and business
when dynamic matrices are partially unknown. Similarly, knowledge are often vague, imprecise, and plagued with un-
Zhang et al. [232] proposed a novel parallel tracking control certainty. And, further, these issues can occur across the entire
optimization algorithm using fuzzy RL techniques for partially recommendation process from collecting the data to generating
unknown fuzzy interconnected systems. This algorithm uses the recommendations. Other key problems that can occur with
the pre-compensation technique to treat working feedback recommender systems include sparsely populated user-item
controls as reconstructed dynamics with virtual controls. This matrices and problems with measuring the similarity of items
approach to building the model results in a new augmented and users (see Fig. 3). Commonly used fuzzy techniques to
and interconnected fuzzy tracking system where a valid per- deal with these issues include intuitionistic fuzzy sets [241],
formance index is guaranteed for optimal control. In the realm fuzzy user profiles [242], fuzzy rule-based systems [243], and
of traffic light control systems, Kumar et al. [233] proposed a fuzzy similarity [244]. This section provides a summary of
novel system that is both dynamic and intelligent to overcome recent articles focused on these techniques.
issues with long waiting times, fuel waste, and rising carbon Collaborative filtering is a key approach to recommender
emissions. Traditional traffic light systems operate on a fixed systems. However, traditional collaborative filtering methods,
duration mode, whereas Kumar’s proposed system uses a such as unsupervised clustering, are quite sensitive to un-
deep RL model to switch the lights and a fuzzy inference certainty and therefore often experience high error rates.
system to select one among three modes based on current Once again, FCM or modified FCM algorithms have been
traffic information. To mitigate frequency deviations caused implemented to eliminate these issues [245]–[247]. For ex-
by power fluctuations, Yin et al. [239] developed a fuzzy ample, FCM has been used to classify the users in a dataset
vector RL approach to control how much power a power according to the similarity of their item ratings. To improve the
system generates. The framework also considers flywheel quality of recommender systems with sparse datasets, Nilashi
energy storage systems. et al. [248] designed a hybrid item similarity model that
Turning to large-scale multi-agent RL, Li et al. [234] combines an adjusted Google similarity with an intuitionistic
introduced the concept of fuzzy agents to be used for training Kullback–Leibler similarity based on fuzzy sets. This approach
homogeneous agents. They also proposed a new RL method essentially makes a trade-off between prediction accuracy and
that uses fuzzy logic to learn abstract policies. In comparison efficiency.
to other simplification methods, their fuzzy agents both reduce In terms of content-based filtering recommender systems,
the computing resources required to train a model and ensure Yera et al.’s [249] solution uses a fuzzy decision tree to match
that an effective policy is learned. Zhu et al. [235] devised the most appropriate function in the individual recommenda-
a new control strategy based on RL and a fuzzy wavelet tion aggregation step. Other researchers have also relied on
network. The aim here is to improve the stability of the hybrid fuzzy rule-based systems to extract relevant knowledge from
system’s buffer compliance control. To reduce the negative uncertain data to improve the performance of knowledge-based
effects of noisy information in communication channels on recommender systems [250]–[252].
multi-agent RL, Fang et al. [236] developed a two-stream Hybrid recommender systems are another area of research
fused fuzzy deep neural network by applying fuzzy a inference progress. Here, Walek et al. [253] designed a new hybrid
module and a DNN module. Experiments with two large-scale recommender system that combines a collaborative filtering

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

11

TABLE V
S UMMARY OF THE SOTA FUZZY TECHNIQUES - BASED RECOMMENDER SYSTEMS B. Fuzzy transfer learning
ACIEVEMENTS .
Recent fuzzy transfer learning works [191], [277] mainly
focus on applying fuzzy rule-based systems to model the
Fuzzy techniques Type No.
uncertainty and variability between different source and target
Collaborative Content-Based Knowledge-Based Hybrid
domains, enabling more effective adaptation of knowledge
Fuzzy systems [248] [254] [255] [256] [257] [250] [251] [261] [262] 12
[258] [259] [260] from the source to the target domain. However, open-set
Intuitionistic [263] [264] - - - 2 problems [165], [278] has gain more and more attention in
fuzzy set transfer learning, where target domain contain private cate-
Fuzzy clustering [245] [246] [247] - - - 3 gories. Detecting unknown classes is an challenge problem
Fuzzy profile [265] [266] - - - 2 that contain a large degree of uncertainty. We believe applying
Others [267] [265] [268] [249] [269] [270] [271] [273] [274] 13 fuzzy techniques to address this challenge problem is worth
[271] [272] [275] [253]
investigating for future work.
system, a content-based filtering system, and a fuzzy expert
system to enhance recommendation performance. The fuzzy C. Fuzzy data stream learning
expert system is used to evaluate the importance of the A couple of recent works [222], [224] in fuzzy data stream
recommended products with vague information and rank them learning are focused on developing adaptive fuzzy models that
appropriately for users. Table V summarizes these recent can effectively handle concept drift in data streams. Learning
studies in the field of fuzzy-based recommender systems. from multiple stream [20] is a crucial and challenge problem
In summary, fuzzy techniques provide a rich spectrum of in data stream learning, especially when streams have different
methods for managing uncertainty, vagueness, and imprecision rates, arrive asynchronously, or experience delays. Streams
in data both during the learning process and when mak- may vary in terms of data types, formats, and modalities.
ing recommendations. Particularly, fuzzy techniques are well Additionally, there is an uncertain relationship between each
suited to handling imprecise user preference descriptions (e.g. pair of streams. Traditional machine learning algorithms face
in linguistic terms), knowledge description, and the gradual difficulty in addressing these challenges. Therefore, we recom-
accumulation of user preference profiles. Therefore, applying mend that researchers use fuzzy techniques in future studies
fuzzy techniques in recommender system can bring more to tackle these issues.
efficient and accurate performance than non-fuzzy models.
D. Fuzzy reinforcement learning
VIII. F UTURE R ESEARCH D IRECTIONS Recent research [233], [237] is mainly focus on integrating
fuzzy systems with RL for improved performance in complex
So far we have summarized recent achievements of FML. In
and dynamic environments. Research is exploring the inte-
this section, we aim to give further discussion of FML’s current
gration of fuzzy logic into Q-learning algorithms to handle
research trends and share some insights on future research
uncertainties in estimating state-action values. Further, fuzzy
directions.
logic is applied to model and handle uncertain or imprecise
reward signals in RL. However, in multi-agent RL [234],
A. Fuzzy classic machine learning capturing complex relationships and dependencies between
Most current research in fuzzy classic machine learning agents while maintaining a scalable and efficient learning pro-
is mainly focusing on the following aspects: 1) handling cess is a key challenge. Additionally, agents in a multi-agent
noisy or incomplete data; 2) addressing imbalanced datasets; system may have diverse capabilities, objectives, or learning
and 3) enhancing algorithms’ interpretability and robustness. speeds. Coordinating heterogeneous agents and ensuring fair
Analyzing imprecise data (fuzzy-valued or interval-valued) and effective collaboration is another challenging problem. We
[84] has not received widespread attention. However, in many believe it would be a promising direction to investigate how
real-world scenarios, we will inevitably encounter this kind to apply fuzzy techniques to address these challenges.
of data. Therefore, it would be a promising direction to
investigate how to apply fuzzy logic to analyze imprecise data. E. Fuzzy recommender systems
Moreover, deep learning [86] has made significant strides, Fuzzy techniques are mainly used to handling imprecise
but it still faces several challenges. Deep neural networks, user preference descriptions (e.g. in linguistic terms), knowl-
particularly complex architectures like deep convolutional or edge description, and the gradual accumulation of user pref-
recurrent networks, are often viewed as black boxes. Un- erence profiles in fuzzy recommender systems [260], [264].
derstanding how these models arrive at specific decisions Cross-domain recommendations [279], where recommenda-
is crucial, especially in applications where interpretability is tions are made across different domains or platforms, present
essential, such as healthcare and finance. In addition, deep several challenges due to the diversity and heterogeneity
learning models are vulnerable to adversarial attacks [276], between domains. Different domains may have distinct char-
where small, carefully crafted perturbations to input data can acteristics, user behaviors, and item features. Moreover, there
lead to misclassification. Fuzzy techniques are potential tools will be uncertain relationships between different domains.
to overcome these challenges. We suggest future work that Applying fuzzy techniques to address these challenges in
uses fuzzy techniques to overcome these challenges. cross-domain recommendations is promising in future work.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

12

IX. S UMMARY 15 Pan, S. J. and Yang, Q., “A survey on transfer learning,” IEEE Transac-
tions on Knowledge and Data Engineering, vol. 22, no. 10, pp. 1345–
In this paper, we reviewed recent developments across the 1359, 2010.
five main research streams of FML. Our review shows that 16 Lu, J., Zuo, H., and Zhang, G., “Fuzzy multiple-source transfer learning,”
fuzzy techniques can significantly improve machine learning IEEE Transactions on Fuzzy Systems, vol. 28, no. 12, pp. 3418–3431,
2020.
algorithms by providing a way to handle different uncertainty 17 Liu, Feng and Zhang, Guangquan and Lu, Jie, “Multi-source het-
situations. The main improvements are reflected in the fol- erogeneous unsupervised domain adaptation via fuzzy-relation neural
lowing five aspects: 1) enhancing the representation of the networks,” IEEE Transactions on Fuzzy Systems, 2020.
18 Liu, F., Lu, J., and Zhang, G., “Unsupervised heterogeneous domain
inputs; 2) improving the learning process of different machine adaptation via shared fuzzy equivalence relations,” IEEE Transactions
learning algorithms; 3) enhancing measurement accuracy and on Fuzzy Systems, vol. 26, no. 6, pp. 3555–3568, 2018.
reliability; 4) improving the accuracy of the matching function; 19 Sun, Y., Tang, K., Minku, L. L., Wang, S., and Yao, X., “Online
ensemble learning of data streams with gradually evolved classes,” IEEE
5) enhancing the performance (e.g., accuracy, robustness and Transactions on Knowledge and Data Engineering, vol. 28, no. 6, pp.
interpretability) of the output results. 1532–1545, 2016.
In future research, several new directions in the field of FML 20 Zhou, M., Lu, J., Song, Y., and Zhang, G., “Multi-stream concept
warrant thorough consideration. For instance, applying fuzzy drift self-adaptation using graph neural network,” IEEE Transactions on
Knowledge and Data Engineering, 2023.
techniques to address open-set transfer learning problems, 21 Schlimmer, J. C. and Granger, R. H., “Incremental learning from noisy
where the target domain encompasses classes that are unknown data,” Machine learning, vol. 1, pp. 317–354, 1986.
in the source domain. In addition, multi-stream learning, 22 Gama, J., Žliobaitė, I., Bifet, A., Pechenizkiy, M., and Bouchachia, A.,
“A survey on concept drift adaptation,” ACM computing surveys, vol. 46,
multi-agent RL, and cross-domain recommendations are three no. 4, pp. 1–37, 2014.
challenge problems that are far from being solved. They all 23 Lu, J., Liu, A., Dong, F., Gu, F., Gama, J., and Zhang, G., “Learning
involve intricate relationships and heterogeneous information, under concept drift: A review,” IEEE transactions on knowledge and
data engineering, vol. 31, no. 12, pp. 2346–2363, 2018.
posing difficulties for traditional machine learning algorithms. 24 Song, Y., Lu, J., Lu, H., and Zhang, G., “Fuzzy clustering-based
Fuzzy techniques emerge as promising tools for investigating adaptive regression for drifting data streams,” IEEE Transactions on
and addressing these complex problems. Fuzzy Systems, vol. 28, no. 3, pp. 544–557, 2020.
25 Dunn, J. C., “A fuzzy relative of the isodata process and its use
We believe that this survey can provide researchers with in detecting compact well-separated clusters,” Journal of Cybernetics,
SOTA knowledge on machine learning based on fuzzy tech- vol. 3, no. 3, pp. 32–57, 1973.
niques and give a guide on future research directions in the 26 Yang, X., Zhang, G., Lu, J., and Ma, J., “A kernel fuzzy c-means
clustering-based fuzzy support vector machine algorithm for classifi-
field of FML. cation problems with outliers or noises,” IEEE Transactions on Fuzzy
Systems, vol. 19, no. 1, pp. 105–115, 2010.
R EFERENCES 27 Sutton, R. S. and Barto, A. G., Reinforcement learning: An introduction.
MIT press, 2018.
1 Dubois, D. J., Fuzzy sets and systems: theory and applications. Aca-
28 Kormushev, P., Calinon, S., and Caldwell, D. G., “Reinforcement learn-
demic press, 1980, vol. 144.
ing in robotics: Applications and real-world challenges,” Robotics, vol. 2,
2 Wu, H. C., “Probability density functions of fuzzy random variables,”
no. 3, pp. 122–148, 2013.
Fuzzy Sets and Systems, vol. 105, no. 1, pp. 139–158, 1999.
3 Takagi, T. and Sugeno, M., “Fuzzy identification of systems and its 29 Goldwaser, A. and Thielscher, M., “Deep reinforcement learning for
applications to modeling and control,” IEEE Transactions on Systems, general game playing,” in Proceedings of the AAAI Conference on
Man, and Cybernetics, vol. 15, no. 1, pp. 116–132, 1985. Artificial Intelligence, vol. 34, no. 02, 2020, pp. 1701–1708.
4 Lin, Zhu, Fulai, Chung, Shitong, and Wang, “Generalized fuzzy c-means 30 Kiran, B. R., Sobh, I., Talpaert, V., Mannion, P., Al Sallab, A. A., Yo-
clustering algorithm with improved fuzzy partitions.” IEEE Transactions gamani, S., and Pérez, P., “Deep reinforcement learning for autonomous
on Systems, Man Cybernetics: Part B, vol. 39, no. 3, pp. 578–591, 2009. driving: A survey,” IEEE Transactions on Intelligent Transportation
5 Czogala, E., Drewniak, J., and Pedrycz, W., “Fuzzy relation equations Systems, vol. 23, no. 6, pp. 4909–4926, 2021.
on a finite set,” Fuzzy Sets and systems, vol. 7, no. 1, pp. 89–101, 1982. 31 Er, M. J. and Deng, C., “Online tuning of fuzzy inference systems using
6 Sugeno, M., “Fuzzy measure and fuzzy integral,” Transactions of the dynamic fuzzy q-learning,” IEEE Transactions on Systems, Man, and
Society of Instrument and Control Engineers, vol. 8, no. 2, pp. 218–226, Cybernetics, Part B (Cybernetics), vol. 34, no. 3, pp. 1478–1489, 2004.
1972. 32 Ma, X., Sun, Z., and He, Y., “Analysis and design of fuzzy controller
7 Wen, W., Li, J., Lin, S., Chen, J., and Chang, S., “A fuzzy-matching and fuzzy observer,” IEEE Transactions on fuzzy systems, vol. 6, no. 1,
model with grid reduction for lithography hotspot detection,” IEEE pp. 41–51, 1998.
Transactions on Computer-Aided Design of Integrated Circuits and 33 Jouffe, L., “Fuzzy inference system learning by reinforcement methods,”
Systems, vol. 33, no. 11, pp. 1671–1680, 2014. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applica-
8 Luhandjula, M., “Fuzzy optimization: an appraisal,” Fuzzy sets and tions and Reviews), vol. 28, no. 3, pp. 338–355, 1998.
systems, vol. 30, no. 3, pp. 257–282, 1989. 34 Glorennec, P. Y. and Jouffe, L., “Fuzzy q-learning,” in Proceedings of
9 Zadeh, L. A., “Fuzzy sets,” Information and Control, vol. 8, no. 3, pp. 6th international fuzzy systems conference, vol. 2, 1997, pp. 659–662.
338–353, 1965. 35 Bobadilla, J., Ortega, F., Hernando, A., and Gutiérrez, A., “Recom-
10 Zimmermann, H., Fuzzy set theory—and its applications. Springer mender systems survey,” Knowledge-based systems, vol. 46, pp. 109–
Science & Business Media, 2011. 132, 2013.
11 Ma, G., Lu, J., Liu, F., Fang, Z., and Zhang, G., “Multiclass classifi- 36 Lu, J., Wu, D., Mao, M., Wang, W., and Zhang, G., “Recommender
cation with fuzzy-feature observations: Theory and algorithms,” IEEE system application developments: a survey,” Decision Support Systems,
Transactions on Cybernetics, vol. 54, no. 2, pp. 1048–1061, 2022. vol. 74, pp. 12–32, 2015.
12 Baraldi, A. and Blonda, P., “A survey of fuzzy clustering algorithms 37 Yager, R. R., “Fuzzy logic methods in recommender systems,” Fuzzy
for pattern recognition. i,” IEEE Transactions on Systems, Man, and Sets and Systems, vol. 136, no. 2, pp. 133–149, 2003.
Cybernetics, Part B (Cybernetics), vol. 29, no. 6, pp. 778–785, 1999. 38 Yera, R. and Martinez, L., “Fuzzy tools in recommender systems: A
13 Škrjanc, I., Iglesias, J. A., Sanchis, A., Leite, D., Lughofer, E., and survey,” International Journal of Computational Intelligence Systems,
Gomide, F., “Evolving fuzzy and neuro-fuzzy approaches in clustering, vol. 10, no. 1, p. 776, 2017.
regression, identification, and classification: A survey,” Information Sci- 39 Yang, M., “A survey of fuzzy clustering,” Mathematical and Computer
ences, vol. 490, pp. 344–368, 2019. Modelling, vol. 18, no. 11, pp. 1–16, 1993.
14 Zheng, Y., Xu, Z., and Wang, X., “The fusion of deep learning and 40 Ruspini, E. H., Bezdek, J. C., and Keller, J. M., “Fuzzy clustering:
fuzzy systems: A state-of-the-art survey,” IEEE Transactions on Fuzzy A historical perspective,” IEEE Computational Intelligence Magazine,
Systems, vol. 30, no. 8, pp. 2783–2799, 2021. vol. 14, no. 1, pp. 45–55, 2019.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

13

41 Bezdek, J. C., Ehrlich, R., and Full, W., “Fcm: The fuzzy c-means 67 Chachi, J., “A weighted least squares fuzzy regression for crisp input-
clustering algorithm,” Computers & Geosciences, vol. 10, no. 2-3, pp. fuzzy output data,” IEEE Transactions on Fuzzy Systems, vol. 27, no. 4,
191–203, 1984. pp. 739–748, 2018.
42 Ding, Y. and Fu, X., “Kernel-based fuzzy c-means clustering algorithm 68 Choi, S. H., Jung, H., and Kim, H., “Ridge fuzzy regression model,”
based on genetic algorithm,” Neurocomputing, vol. 188, pp. 233–238, International Journal of Fuzzy Systems, vol. 21, no. 7, pp. 2077–2090,
2016. 2019.
43 Lei, T., Jia, X., Zhang, Y., He, L., Meng, H., and Nandi, A. K., 69 Naderkhani, R., Behzad, M., Razzaghnia, T., and Farnoosh, R., “Fuzzy
“Significantly fast and robust fuzzy c-means clustering algorithm based regression analysis based on fuzzy neural networks using trapezoidal
on morphological reconstruction and membership filtering,” IEEE Trans- data,” International Journal of Fuzzy Systems, vol. 23, pp. 1267–1280,
actions on Fuzzy Systems, vol. 26, no. 5, pp. 3027–3041, 2018. 2021.
44 Mishro, P. K., Agrawal, S., Panda, R., and Abraham, A., “A novel type- 70 Xia, H., Tang, J., Yu, W., Cui, C., and Qiao, J., “Takagi–sugeno fuzzy
2 fuzzy c-means clustering for brain mr image segmentation,” IEEE regression trees with application to complex industrial modeling,” IEEE
Transactions on Cybernetics, vol. 51, no. 8, pp. 3901–3912, 2020. Transactions on Fuzzy Systems, 2022.
45 Zhou, J., Pedrycz, W., Yue, X., Gao, C., Lai, Z., and Wan, J., “Projected 71 Zhang, D., Xu, Y., Peng, Y., Du, C., Wang, N., Tang, M., Lu, L., and
fuzzy c-means clustering with locality preservation,” Pattern Recogni- Liu, J., “An interpretable station delay prediction model based on graph
tion, vol. 113, p. 107748, 2021. community neural network and time-series fuzzy decision tree,” IEEE
46 Gao, Y., Wang, Z., Xie, J., and Pan, J., “A new robust fuzzy c-means Transactions on Fuzzy Systems, 2022.
clustering method based on adaptive elastic distance,” Knowledge-Based 72 Yuan, Y. and Shaw, M. J., “Induction of fuzzy decision trees,” Fuzzy
Systems, vol. 237, p. 107769, 2022. Sets and Systems, vol. 69, no. 2, pp. 125–139, 1995.
47 Lei, T., Liu, P., Jia, X., Zhang, X., Meng, H., and Nandi, A. K., 73 Suárez, A. and Lutsko, J. F., “Globally optimal fuzzy decision trees for
“Automatic fuzzy clustering framework for image segmentation,” IEEE classification and regression,” IEEE Transactions on Pattern Analysis
Transactions on Fuzzy Systems, vol. 28, no. 9, pp. 2078–2092, 2019. and Machine Intelligence, vol. 21, no. 12, pp. 1297–1311, 1999.
48 Jiao, L., Yang, H., Liu, Z., and Pan, Q., “Interpretable fuzzy clustering 74 Chiang, I. and Hsu, J. Y., “Fuzzy classification trees for data analysis,”
using unsupervised fuzzy decision trees,” Information Sciences, vol. 611, Fuzzy Sets and Systems, vol. 130, no. 1, pp. 87–99, 2002.
pp. 540–563, 2022. 75 Wang, X., Liu, X., Pedrycz, W., and Zhang, L., “Fuzzy rule based
49 Sangma, J. W., Pal, V., Kumar, N., Kushwaha, R. et al., “Fhc-nds: decision trees,” Pattern Recognition, vol. 48, no. 1, pp. 50–59, 2015.
Fuzzy hierarchical clustering of multiple nominal data streams,” IEEE 76 Segatori, A., Marcelloni, F., and Pedrycz, W., “On distributed fuzzy
Transactions on Fuzzy Systems, vol. 31, no. 3, pp. 786–798, 2022. decision trees for big data,” IEEE Transactions on Fuzzy Systems, vol. 26,
50 Chukhrova, N. and Johannssen, A., “Fuzzy regression analysis: sys- no. 1, pp. 174–192, 2017.
tematic review and bibliography,” Applied Soft Computing, vol. 84, p. 77 Nauck, D. and Kruse, R., “A neuro-fuzzy method to learn fuzzy
105708, 2019. classification rules from data,” Fuzzy sets and Systems, vol. 89, no. 3,
51 Zadeh, L. A., “The concept of a linguistic variable and its application pp. 277–288, 1997.
to approximate reasoning—i,” Information Sciences, vol. 8, no. 3, pp. 78 Cpalka, K., “A new method for design and reduction of neuro-fuzzy
199–249, 1975. classification systems,” IEEE Transactions on Neural Networks, vol. 20,
52 Hong, D. H. and Hwang, C., “Support vector fuzzy regression machines,” no. 4, pp. 701–714, 2009.
Fuzzy Sets and Systems, vol. 138, no. 2, pp. 271–281, 2003. 79 Ghosh, A., Shankar, B. U., and Meher, S. K., “A novel approach to
53 Lu, J., Yang, X., and Zhang, G., “Support vector machine based approach neuro-fuzzy classification,” Neural Networks, vol. 22, no. 1, pp. 100–
in situation assessment,” Dynamics of Continuous Discrete and Impulsive 109, 2009.
Systems-series B-applications & Algorithms, 2006. 80 Juang, C. and Hsieh, C.-D., “Ts-fuzzy system-based support vector
54 Lu, J. and Yang, X. and Zhang, G., “Support vector machine-based multi- regression,” Fuzzy Sets and Systems, vol. 160, no. 17, pp. 2486–2504,
source multi-attribute information integration for situation assessment,” 2009.
Expert Systems with Applications, vol. 34, no. 2, pp. 1333–1340, 2008. 81 Chen, L., Wu, M., Zhou, M., Liu, Z., She, J., and Hirota, K., “Dynamic
55 Dunyak, J. P. and Wunsch, D., “Fuzzy regression by fuzzy number neural emotion understanding in human–robot interaction based on two-layer
networks,” Fuzzy Sets and Systems, vol. 112, no. 3, pp. 371–380, 2000. fuzzy svr-ts model,” IEEE Transactions on Systems, Man, and Cyber-
56 Zhang, D., Deng, L., Cai, K., and So, A., “Fuzzy nonlinear regression netics: Systems, vol. 50, no. 2, pp. 490–501, 2017.
with fuzzified radial basis function network,” IEEE Transactions on 82 Rabcan, J., Levashenko, V., Zaitseva, E., Kvassay, M., and Subbotin, S.,
Fuzzy Systems, vol. 13, no. 6, pp. 742–760, 2005. “Application of fuzzy decision tree for signal classification,” IEEE
57 Mosleh, M., Otadi, M., and Abbasbandy, S., “Fuzzy polynomial re- Transactions on Industrial Informatics, vol. 15, no. 10, pp. 5425–5434,
gression with fuzzy neural networks,” Applied Mathematical Modelling, 2019.
vol. 35, no. 11, pp. 5400–5412, 2011. 83 Xue, G., Chang, Q., Wang, J., Zhang, K., and Pal, N. R., “An adaptive
58 Roh, S., Ahn, T., and Pedrycz, W., “Fuzzy linear regression based on neuro-fuzzy system with integrated feature selection and rule extraction
polynomial neural networks,” Expert Systems with Applications, vol. 39, for high-dimensional classification problems,” IEEE Transactions on
no. 10, pp. 8909–8928, 2012. Fuzzy Systems, 2022.
59 Tanaka, H. and Lee, H., “Interval regression analysis by quadratic 84 Ma, G., Liu, F., Zhang, G., and Lu, J., “Learning from imprecise obser-
programming approach,” IEEE Transactions on Fuzzy Systems, vol. 6, vations: An estimation error bound based on fuzzy random variables,” in
no. 4, pp. 473–481, 1998. 2021 IEEE International Conference on Fuzzy Systems, 2021, pp. 1–8.
60 Hong, D. and Hwang, C., “Interval regression analysis using quadratic 85 Liu, P., Han, S., Meng, Z., and Tong, Y., “Facial expression recognition
loss support vector machine,” IEEE Transactions on Fuzzy Systems, via a boosted deep belief network,” in Proceedings of the IEEE Confer-
vol. 13, no. 2, pp. 229–237, 2005. ence on Computer Vision and Pattern Recognition, 2014, pp. 1805–1812.
61 Hwang, C., Hong, D., and Seok, K., “Support vector interval regression 86 Gu, J., Wang, Z., Kuen, J., Ma, L., Shahroudy, A., Shuai, B., Liu, T.,
machine for crisp input and output data,” Fuzzy Sets and Systems, vol. Wang, X., Wang, G., Cai, J. et al., “Recent advances in convolutional
157, no. 8, pp. 1114–1125, 2006. neural networks,” Pattern Recognition, vol. 77, pp. 354–377, 2018.
62 Hao, P., “Interval regression analysis using support vector networks,” 87 Schuster, M. and Paliwal, K. K., “Bidirectional recurrent neural net-
Fuzzy Sets and Systems, vol. 160, no. 17, pp. 2466–2485, 2009. works,” IEEE transactions on Signal Processing, vol. 45, no. 11, pp.
63 Huang, C., “A reduced support vector machine approach for interval 2673–2681, 1997.
regression analysis,” Information Sciences, vol. 217, pp. 56–64, 2012. 88 Shihabudheen, K. and Pillai, G. N., “Recent advances in neuro-fuzzy
64 He, Y., Wang, X., and Huang, J. Z., “Fuzzy nonlinear regression analysis system: A survey,” Knowledge-Based Systems, vol. 152, pp. 136–162,
using a random weight network,” Information Sciences, vol. 364, pp. 2018.
222–240, 2016. 89 De Campos Souza, P. V., “Fuzzy neural networks and neuro-fuzzy
65 He, Y., Wei, C., Long, H., Ashfaq, R. A. R., and Huang, J. Z., “Random networks: A review the main techniques and applications used in the
weight network-based fuzzy nonlinear regression for trapezoidal fuzzy literature,” Applied Soft Computing, vol. 92, p. 106275, 2020.
number data,” Applied Soft Computing, vol. 70, pp. 959–979, 2018. 90 Chen, L., Su, W., Wu, M., Pedrycz, W., and Hirota, K., “A fuzzy
66 Baser, F. and Demirhan, H., “A fuzzy regression with support vector deep neural network with sparse autoencoder for emotional intention
machine approach to the estimation of horizontal global solar radiation,” understanding in human–robot interaction,” IEEE Transactions on Fuzzy
Energy, vol. 123, pp. 229–240, 2017. Systems, vol. 28, no. 7, pp. 1252–1264, 2020.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

14

91 Lu, H., Zhang, M., Xu, X., Li, Y., and Shen, H. T., “Deep fuzzy logic for images scoring,” IEEE Transactions on Fuzzy Systems, vol. 29,
hashing network for efficient image retrieval,” IEEE Transactions on no. 1, pp. 34–45, 2020.
Fuzzy Systems, vol. 29, no. 1, pp. 166–176, 2020. 112 Cunha Sergio, G. and Lee, M., “Emotional video to audio transforma-
92 Cao, B., Zhao, J., Lv, Z., Gu, Y., Yang, P., and Halgamuge, S. K., tion using deep recurrent neural networks and a neuro-fuzzy system,”
“Multiobjective evolution of fuzzy rough neural network via distributed Mathematical Problems in Engineering, vol. 2020, pp. 1–15, 2020.
parallelism for stock prediction,” IEEE Transactions on Fuzzy Systems, 113 Wang, G., Zhou, T., Choi, K., and Lu, J., “A deep-ensemble-level-based
vol. 28, no. 5, pp. 939–952, 2020. interpretable takagi–sugeno–kang fuzzy classifier for imbalanced data,”
93 Cao, B., Zhao, J., Liu, X., Arabas, J., Tanveer, M., Singh, A. K., and IEEE Transactions on Cybernetics, vol. 52, no. 5, pp. 3805–3818, 2020.
Lv, Z., “Multiobjective evolution of the explainable fuzzy rough neural 114 Ghasemi, M., Kelarestaghi, M., Eshghi, F., and Sharifi, A., “T2-fdl:
network with gene expression programming,” IEEE Transactions on a robust sparse representation method using adaptive type-2 fuzzy
Fuzzy Systems, vol. 30, no. 10, pp. 4190–4200, 2022. dictionary learning for medical image classification,” Expert Systems with
94 Sideratos, G., Ikonomopoulos, A., and Hatziargyriou, N. D., “A novel Applications, vol. 158, p. 113500, 2020.
fuzzy-based ensemble model for load forecasting using hybrid deep 115 Sharma, T. and Verma, N. K., “Estimating depth and global atmospheric
neural networks,” Electric Power Systems Research, vol. 178, p. 106025, light for image dehazing using type-2 fuzzy approach,” IEEE Transac-
2020. tions on Emerging Topics in Computational Intelligence, vol. 6, no. 1,
95 Banerjee, A., Singh, P. K., and Sarkar, R., “Fuzzy integral-based cnn pp. 93–102, 2020.
classifier fusion for 3d skeleton action recognition,” IEEE Transactions
116 Asadi, R. and Regan, A. C., “A spatio-temporal decomposition based
on Circuits and Systems for Video Technology, vol. 31, no. 6, pp. 2206–
deep neural network for time series forecasting,” Applied Soft Computing,
2216, 2020.
vol. 87, p. 105963, 2020.
96 Hsu, M., Chien, Y., Wang, W., and Hsu, C., “A convolutional fuzzy
neural network architecture for object classification with small training 117 Han, H., Zhang, H., and Qiao, J., “Robust deep neural network using
database,” International Journal of Fuzzy Systems, vol. 22, pp. 1–10, fuzzy denoising autoencoder,” International Journal of Fuzzy Systems,
2020. vol. 22, pp. 1356–1375, 2020.
97 Ge, C., Liu, Z., Fang, L., Ling, H., Zhang, A., and Yin, C., “A hybrid 118 Altameem, T., “Fuzzy rank correlation-based segmentation method and
fuzzy convolutional neural network based mechanism for photovoltaic deep neural network for bone cancer identification,” Neural Computing
cell defect detection with electroluminescence images,” IEEE Transac- and Applications, vol. 32, pp. 805–815, 2020.
tions on Parallel and Distributed Systems, vol. 32, no. 7, pp. 1653–1664, 119 Wang, F., Pedrycz, W., Herrera, F., and Su, S., “Fuzzy logic and
2020. artificial intelligence: A special issue on emerging techniques and their
98 Sharma, T., Verma, N. K., and Masood, S., “Mixed fuzzy pooling applications,” IEEE Transactions on Fuzzy Systems, vol. 28, no. 12, pp.
in convolutional neural networks for image classification,” Multimedia 3063–3064, 2020.
Tools and Applications, vol. 82, no. 6, pp. 8405–8421, 2023. 120 Reddy, T. K., Arora, V., Gupta, V., Biswas, R., and Behera, L., “Eeg-
99 Bhalla, K., Koundal, D., Sharma, B., Hu, Y., and Zaguia, A., “A fuzzy based drowsiness detection with fuzzy independent phase-locking value
convolutional neural network for enhancing multi-focus image fusion,” representations using lagrangian-based deep neural networks,” IEEE
Journal of Visual Communication and Image Representation, vol. 84, p. Transactions on Systems, Man, and Cybernetics: Systems, vol. 52, no. 1,
103485, 2022. pp. 101–111, 2021.
100 Bhalla, K., Koundal, D., Bhatia, S., Rahmani, M. K. I., and Tahir, M., 121 Alzubi, J. A., Jain, R., Nagrath, P., Satapathy, S., Taneja, S., and
“Fusion of infrared and visible images using fuzzy based siamese Gupta, P., “Deep image captioning using an ensemble of cnn and lstm
convolutional network,” Computers, Materials & Continua, vol. 70, based deep neural networks,” Journal of Intelligent & Fuzzy Systems,
no. 3, pp. 5503–5518, 2022. vol. 40, no. 4, pp. 5761–5769, 2021.
101 Khatter, H. and Ahlawat, A. K., “An intelligent personalized web 122 Asghar, M. Z., Subhan, F., Ahmad, H., Khan, W. Z., Hakak, S.,
blog searching technique using fuzzy-based feedback recurrent neural Gadekallu, T. R., and Alazab, M., “Senti-esystem: a sentiment-based
network,” Soft Computing, vol. 24, pp. 9321–9333, 2020. esystem-using hybridized fuzzy and deep neural network for measuring
102 Concepción, L., Nápoles, G., Grau, I., and Pedrycz, W., “Fuzzy-rough customer satisfaction,” Software: Practice and Experience, vol. 51, no. 3,
cognitive networks: Theoretical analysis and simpler models,” IEEE pp. 571–594, 2021.
Transactions on Cybernetics, vol. 52, no. 5, pp. 2994–3005, 2020. 123 Wang, G. and Qiao, J., “An efficient self-organizing deep fuzzy neural
103 Li, R., Hu, Y., and Liang, Q., “T2f-lstm method for long-term traffic network for nonlinear system modeling,” IEEE Transactions on Fuzzy
volume prediction,” IEEE Transactions on Fuzzy Systems, vol. 28, no. 12, Systems, vol. 30, no. 7, pp. 2170–2182, 2021.
pp. 3256–3264, 2020. 124 Pourmeidani, H. and Demara, R. F., “High-accuracy deep belief network:
104 Bilgili, M., Yildirim, A., Ozbek, A., Celebi, K., and Ekinci, F., “Long Fuzzy neural networks using mram-based stochastic neurons,” IEEE
short-term memory (lstm) neural network and adaptive neuro-fuzzy Journal on Exploratory Solid-State Computational Devices and Circuits,
inference system (anfis) approach in modeling renewable electricity vol. 7, no. 2, pp. 125–131, 2021.
generation forecasting,” International Journal of Green Energy, vol. 18, 125 Sharma, R., Goel, T., Tanveer, M., Dwivedi, S., and Murugan, R., “Faf-
no. 6, pp. 578–594, 2021. drvfl: Fuzzy activation function based deep random vector functional
105 Xue, M., Yan, H., Wang, M., Shen, H., and Shi, K., “Lstm-based intel- links network for early diagnosis of alzheimer disease,” Applied Soft
ligent fault detection for fuzzy markov jump systems and its application Computing, vol. 106, p. 107371, 2021.
to tunnel diode circuits,” IEEE Transactions on Circuits and Systems II:
126 Qasem, S. N. and Mohammadzadeh, A., “A deep learned type-2 fuzzy
Express Briefs, vol. 69, no. 3, pp. 1099–1103, 2021.
neural network: Singular value decomposition approach,” Applied Soft
106 Li, Y., Tong, Z., Tong, S., and Westerdahl, D., “A data-driven interval
Computing, vol. 105, p. 107244, 2021.
forecasting model for building energy prediction using attention-based
lstm and fuzzy information granulation,” Sustainable Cities and Society, 127 Wang, Y. and Luo, C., “An intelligent quantitative trading system based
vol. 76, p. 103481, 2022. on intuitionistic-gru fuzzy neural networks,” Applied Soft Computing,
107 Shen, T., Wang, J., Gou, C., and Wang, F., “Hierarchical fused model vol. 108, p. 107471, 2021.
with deep learning and type-2 fuzzy learning for breast cancer diagnosis,” 128 Qazani, M. R. C., Asadi, H., Al-Ashmori, M., Mohamed, S., Lim, C. P.,
IEEE Transactions on Fuzzy Systems, vol. 28, no. 12, pp. 3204–3218, and Nahavandi, S., “Time series prediction of driving motion scenarios
2020. using fuzzy neural networks:* motion signal prediction using fnns,” in
108 Kalaiselvi, S. and Gomathi, V., “α-cut induced fuzzy deep neural 2021 IEEE International Conference on Mechatronics, 2021, pp. 1–6.
network for change detection of sar images,” Applied Soft Computing, 129 Sun, J., Cao, R., Zhou, M., Hussain, W., Wang, B., Xue, J., and Xiang, J.,
vol. 95, p. 106510, 2020. “A hybrid deep neural network for classification of schizophrenia using
109 Özyurt, F., Sert, E., and Avcı, D., “An expert system for brain tumor eeg data,” Scientific Reports, vol. 11, no. 1, pp. 1–16, 2021.
detection: Fuzzy c-means with super resolution and convolutional neural 130 Dong, L., Jiang, F., Wang, M., and Li, X., “Fuzzy deep wavelet
network with extreme learning machine,” Medical Hypotheses, vol. 134, neural network with hybrid learning algorithm: application to electrical
p. 109433, 2020. resistivity imaging inversion,” Knowledge-Based Systems, vol. 242, p.
110 Puchkov, A., Dli, M., and Kireyenkova, M., “Fuzzy classification on the 108164, 2022.
base of convolutional neural networks,” in Advances in Artificial Systems 131 Yu, J., Kim, J., Li, X., Jong, Y., Kim, K., and Ryang, G., “Water quality
for Medicine and Education II 2, 2020, pp. 379–391. forecasting based on data decomposition, fuzzy clustering and deep
111 Kang, C., Yu, X., Wang, S., Guttery, D. S., Pandey, H. M., Tian, Y., learning neural network,” Environmental Pollution, vol. 303, p. 119136,
and Zhang, Y., “A heuristic neural network structure relying on fuzzy 2022.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

15

132 Ghosal, S., Sarkar, M., and Sarkar, R., “Nofed-net: Nonlinear fuzzy network embedding for recommendation,” Information Sciences, vol.
ensemble of deep neural networks for human activity recognition,” IEEE 620, pp. 105–124, 2023.
Internet of Things Journal, vol. 9, no. 18, pp. 17 526–17 535, 2022. 155 Aghaeipoor, F., Sabokrou, M., and Fernández, A., “Fuzzy rule-based
133 He, L., Chen, Y., and Wu, K., “Fuzzy granular deep convolutional explainer systems for deep neural networks: From local explainability to
network with residual structures,” Knowledge-Based Systems, vol. 258, global understanding,” IEEE Transactions on Fuzzy Systems, 2023.
p. 109941, 2022. 156 Zhang, L., Shi, Y., Chang, Y., and Lin, C., “Robust fuzzy neural network
134 Zheng, W., Yan, L., Gou, C., and Wang, F., “Fuzzy deep forest with with an adaptive inference engine,” IEEE Transactions on Cybernetics,
deep contours feature for leaf cultivar classification,” IEEE Transactions 2023.
on Fuzzy Systems, vol. 30, no. 12, pp. 5431–5444, 2022. 157 Sindhuraj, I. C. G. L. and Patrick, A. J., “Loan eligibility prediction
135 Vega, C. F., Quevedo, J., Escandón, E., Kiani, M., Ding, W., and Andreu- using adaptive hybrid optimization driven-deep neuro fuzzy network,”
Perez, J., “Fuzzy temporal convolutional neural networks in p300- Expert Systems with Applications, p. 119903, 2023.
based brain–computer interface for smart home interaction,” Applied Soft 158 Lu, J., Behbood, V., Hao, P., Zuo, H., Xue, S., and Zhang, G., “Transfer
Computing, vol. 117, p. 108359, 2022. learning using computational intelligence: A survey,” Knowledge-Based
136 Kumar, A., Sangwan, S. R., Arora, A., and Menon, V. G., “Depress- Systems, vol. 80, pp. 14–23, 2015.
dcnf: a deep convolutional neuro-fuzzy model for detection of depression 159 Zhang, X., Zhang, X., Liu, H., and Liu, X., “Multi-task clustering
episodes using iomt,” Applied Soft Computing, vol. 122, p. 108863, 2022. through instances transfer,” Neurocomputing, vol. 251, pp. 145–155,
137 Zhou, W., Liu, M., and Xu, Z., “The dual-fuzzy convolutional neural 2017.
network to deal with handwritten image recognition,” IEEE Transactions 160 Venkateswara, H., Chakraborty, S., and Panchanathan, S., “Deep-learning
on Fuzzy Systems, vol. 30, no. 12, pp. 5225–5236, 2022. systems for domain adaptation in computer vision: Learning transferable
138 Zaremarjal, A. Y., Yiltas-Kaplan, D., and Lazemi, S., “Emotion ex- feature representations,” IEEE Signal Processing Magazine, vol. 34,
traction from text using fuzzy-deep neural network,” in Intelligent and no. 6, pp. 117–129, 2017.
Fuzzy Techniques for Emerging Conditions and Digital Transformation: 161 Duan, L., Tsang, I. W., and Xu, D., “Domain transfer multiple kernel
Proceedings of the INFUS 2021 Conference, held August 24-26, 2021. learning,” IEEE Transactions on Pattern Analysis and Machine Intelli-
Volume 2, 2022, pp. 329–338. gence, vol. 34, no. 3, pp. 465–479, 2012.
139 Muthusamy, D. and Rakkimuthu, P., “Trilateral filterative hermitian 162 Dietterich, T. G., Domingos, P., Getoor, L., Muggleton, S., and Tade-
feature transformed deep perceptive fuzzy neural network for finger vein palli, P., “Structured machine learning: the next ten years,” Machine
verification,” Expert Systems with Applications, vol. 196, p. 116678, Learning, vol. 73, no. 1, p. 3, 2008.
2022. 163 Liu, Q., Liao, X., Carin, H. L., Stack, J. R., and Carin, L., “Semisuper-
140 Wang, J., Chang, Q., Gao, T., Zhang, K., and Pal, N. R., “Sensitivity vised multitask learning,” IEEE Transactions on Pattern Analysis and
analysis of takagi–sugeno fuzzy neural network,” Information Sciences, Machine Intelligence, vol. 31, no. 6, pp. 1074–1086, 2009.
vol. 582, pp. 725–749, 2022. 164 Zuo, H., Zhang, G., Behbood, V., Lu, J., and Meng, X., “Transfer
141 Rafiei, H. and Akbarzadeh-T, M., “Reliable fuzzy neural networks for learning in hierarchical feature spaces,” in 2015 10th International
systems identification and control,” IEEE Transactions on Fuzzy Systems, Conference on Intelligent Systems and Knowledge Engineering, 2015,
2022. pp. 183–188.
142 Tu, Q., Zhang, Q., Zhang, Z., Gong, D., and Tang, M., “A deep spa- 165 Fang, Z., Lu, J., Liu, F., Xuan, J., and Zhang, G., “Open set domain
tiotemporal fuzzy neural network for subway passenger flow prediction adaptation: Theoretical bound and algorithm,” IEEE transactions on
with covid-19 search engine data,” IEEE Transactions on Fuzzy Systems, neural networks and learning systems, vol. 32, no. 10, pp. 4309–4322,
vol. 31, no. 2, pp. 394–406, 2022. 2020.
143 Nasiri, H. and Ebadzadeh, M. M., “Mfrfnn: Multi-functional recurrent 166 Yang, J., Yan, R., and Hauptmann, A. G., “Cross-domain video con-
fuzzy neural network for chaotic time series prediction,” Neurocomput- cept detection using adaptive svms,” in Proceedings of the 15th ACM
ing, vol. 507, pp. 292–310, 2022. international conference on Multimedia, 2007, pp. 188–197.
144 Thakare, K. V., Sharma, N., Dogra, D. P., Choi, H., and Kim, I., “A multi- 167 Fang, Z., Lu, J., Liu, F., and Zhang, G., “Semi-supervised heterogeneous
stream deep neural network with late fuzzy fusion for real-world anomaly domain adaptation: Theory and algorithms,” IEEE Transactions on
detection,” Expert Systems with Applications, vol. 201, p. 117030, 2022. Pattern Analysis and Machine Intelligence, vol. 45, no. 1, pp. 1087–
145 Dey, S., Roychoudhury, R., Malakar, S., and Sarkar, R., “An optimized 1105, 2022.
fuzzy ensemble of convolutional neural networks for detecting tubercu- 168 Behbood, V., Lu, J., Zhang, G., and Pedrycz, W., “Multistep fuzzy
losis from chest x-ray images,” Applied Soft Computing, vol. 114, p. bridged refinement domain adaptation algorithm and its application to
108094, 2022. bank failure prediction,” IEEE Transactions on Fuzzy Systems, vol. 23,
146 Tan, D., Huang, Z., Peng, X., Zhong, W., and Mahalec, V., “Deep no. 6, pp. 1917–1935, 2015.
adaptive fuzzy clustering for evolutionary unsupervised representation 169 Wu, D., Lawhern, V. J., Gordon, S., Lance, B. J., and Lin, C., “Driver
learning,” IEEE Transactions on Neural Networks and Learning Systems, drowsiness estimation from eeg signals using online weighted adaptation
2023. regularization for regression (owarr),” IEEE Transactions on Fuzzy
147 Wang, Y., Ishibuchi, H., Er, M. J., and Zhu, J., “Unsupervised multilayer Systems, vol. 25, no. 6, pp. 1522–1535, 2017.
fuzzy neural networks for image clustering,” Information Sciences, vol. 170 Gargees, R., Keller, J. M., and Popescu, M., “Tlpcm: Transfer learning
622, pp. 682–709, 2023. possibilistic c-means,” IEEE Transactions on Fuzzy Systems, 2020.
148 Wang, C., Lv, X., Shao, M., Qian, Y., and Zhang, Y., “A novel fuzzy 171 Sun, S., Yun, J., Lin, H., Zhang, N., Abraham, A., and Liu, H., “Granular
hierarchical fusion attention convolution neural network for medical transfer learning using type-2 fuzzy hmm for text sequence recognition,”
image super-resolution reconstruction,” Information Sciences, vol. 622, Neurocomputing, vol. 214, pp. 126–133, 2016.
pp. 424–436, 2023. 172 Shukla, A. K., Kumar, S., Jagdev, R., Muhuri, P. K., and Lohani, Q. D.,
149 Zhang, C., Lin, Y., Chen, C. P., Yao, H., Cai, H., and Fang, W., “Fuzzy “Interval type-2 fuzzy weighted extreme learning machine for gdp
representation learning on graph,” IEEE Transactions on Fuzzy Systems, prediction,” in 2018 International Joint Conference on Neural Networks,
2023. 2018, pp. 1–8.
150 Ma, X., Chen, L., Deng, Z., Xu, P., Yan, Q., Choi, K., and Wang, S., 173 Kumar, S., Shukla, A. K., Muhuri, P. K., and Lohani, Q. D., “Transfer
“Deep image feature learning with fuzzy rules,” IEEE Transactions on learning based gdp prediction from uncertain carbon emission data,” in
Emerging Topics in Computational Intelligence, 2023. 2019 IEEE International Conference on Fuzzy Systems, 2019, pp. 1–6.
151 Lin, C. and Yang, T., “A fusion-based convolutional fuzzy neural network 174 Shell, J. and Coupland, S., “Fuzzy transfer learning: methodology and
for lung cancer classification,” International Journal of Fuzzy Systems, application,” Information Sciences, vol. 293, pp. 59–79, 2015.
vol. 25, no. 2, pp. 451–467, 2023. 175 Meher, S. K. and Kothari, N. S., “Interpretable rule-based fuzzy elm
152 Raz, N. R., Akbarzadeh-T, M., and Setayeshi, S., “Influence-based and domain adaptation for remote sensing image classification,” IEEE
nano fuzzy swarm oxygen deficiency detection and therapy,” IEEE Transactions on Geoscience and Remote Sensing, 2020.
Transactions on Systems, Man, and Cybernetics: Systems, 2023. 176 Deng, Z., Jiang, Y., Chung, F., Ishibuchi, H., and Wang, S., “Knowledge-
153 Yin, R., Pan, X., Zhang, L., Yang, J., and Lu, W., “A rule-based leverage-based fuzzy system and its modeling,” IEEE Transactions on
deep fuzzy system with nonlinear fuzzy feature transform for data Fuzzy Systems, vol. 21, no. 4, pp. 597–609, 2013.
classification,” Information Sciences, 2023. 177 Deng, Z., Jiang, Y., Choi, K., Chung, F., and Wang, S., “Knowledge-
154 Pham, P., Nguyen, L. T., Nguyen, N. T., Kozma, R., and Vo, B., leverage-based tsk fuzzy system modeling,” IEEE Transactions on Neu-
“A hierarchical fused fuzzy deep neural network with heterogeneous ral Networks and Learning Systems, vol. 24, no. 8, pp. 1200–1212, 2013.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

16

178 Deng, Z., Jiang, Y., Ishibuchi, H., Choi, K., and Wang, S., “Enhanced 201 Zhang, B., Qin, S., Wang, W., Wang, D., and Xue, L., “Data stream
knowledge-leverage-based tsk fuzzy system modeling for inductive trans- clustering based on fuzzy c-mean algorithm and entropy theory,” Signal
fer learning,” ACM Transactions on Intelligent Systems and Technology, Processing, vol. 126, pp. 111–116, 2016.
vol. 8, no. 1, pp. 1–21, 2016. 202 Hong, T., Chen, C., Li, Y., and Wu, M., “Using fuzzy c-means to discover
179 Yang, C., Deng, Z., Choi, K., and Wang, S., “Takagi–sugeno–kang concept-drift patterns for membership functions,” Transactions on Fuzzy
transfer learning fuzzy logic system for the adaptive recognition of Sets and Systems, vol. 1, no. 2, pp. 21–31, 2022.
epileptic electroencephalogram signals,” IEEE Transactions on Fuzzy 203 Song, Y., Zhang, G., Lu, J., and Lu, H., “A fuzzy kernel c-means
Systems, vol. 24, no. 5, pp. 1079–1094, 2016. clustering model for handling concept drift in regression,” in 2017 IEEE
180 Jiang, Y., Wu, D., Deng, Z., Qian, P., Wang, J., Wang, G., Chung, F., International Conference on Fuzzy Systems, 2017, pp. 1–6.
Choi, K., and Wang, S., “Seizure classification from eeg signals using 204 Song, Y., Zhang, G., Lu, H., and Lu, J., “A noise-tolerant fuzzy c-means
transfer learning, semi-supervised learning and tsk fuzzy system,” IEEE based drift adaptation method for data stream regression,” in 2019 IEEE
Transactions on Neural Systems and Rehabilitation Engineering, vol. 25, International Conference on Fuzzy Systems, 2019, pp. 1–6.
no. 12, pp. 2270–2284, 2017. 205 Bechini, A., Marcelloni, F., and Renda, A., “Tsf-dbscan: a novel fuzzy
181 Deng, Z., Xu, P., Xie, L., Choi, K., and Wang, S., “Transductive joint- density-based approach for clustering unbounded data streams,” IEEE
knowledge-transfer tsk fs for recognition of epileptic eeg signals,” IEEE Transactions on Fuzzy Systems, vol. 30, no. 3, pp. 623–637, 2022.
Transactions on Neural Systems and Rehabilitation Engineering, vol. 26, 206 Zhang, Q., Wu, D., Zhang, G., and Lu, J., “Fuzzy user-interest drift
no. 8, pp. 1481–1494, 2018. detection based recommender systems,” in 2016 IEEE International
182 Xie, L., Deng, Z., Xu, P., Choi, K., and Wang, S., “Generalized hidden- Conference on Fuzzy Systems, 2016, pp. 1274–1281.
mapping transductive transfer learning for recognition of epileptic elec- 207 Dong, F., Zhang, G., Lu, J., and Li, K., “Fuzzy competence model drift
troencephalogram signals,” IEEE Transactions on Cybernetics, vol. 49, detection for data-driven decision support systems,” Knowledge-Based
no. 6, pp. 2200–2214, 2018. Systems, vol. 143, pp. 284–294, 2018.
183 Jiang, Y., Gu, X., Ji, D., Qian, P., Xue, J., Zhang, Y., Zhu, J., Xia, K., 208 Pratama, M., Lu, J., Lughofer, E., Zhang, G., and Er, M. J., “An
and Wang, S., “Smart diagnosis: A multiple-source transfer tsk fuzzy incremental learning of concept drifts using evolving type-2 recurrent
system for eeg seizure identification,” ACM Transactions on Multimedia fuzzy neural networks,” IEEE Transactions on Fuzzy Systems, vol. 25,
Computing, Communications, and Applications, vol. 16, no. 2s, pp. 1–21, no. 5, pp. 1175–1192, 2017.
2020. 209 Song, Y., Zhang, G., Lu, H., and Lu, J., “A self-adaptive fuzzy network
184 Jiang, Y., Zhang, Y., Lin, C., Wu, D., and Lin, C., “Eeg-based driver for prediction in non-stationary environments,” in 2018 IEEE Interna-
drowsiness estimation using an online multi-view and transfer tsk fuzzy tional Conference on Fuzzy Systems, 2018, pp. 1–8.
system,” IEEE Transactions on Intelligent Transportation Systems, 2021. 210 Garcia, C., Leite, D., and Škrjanc, I., “Incremental missing-data imputa-
185 Xu, P., Deng, Z., Wang, J., Zhang, Q., Choi, K., and Wang, S., “Transfer tion for evolving fuzzy granular prediction,” IEEE Transactions on Fuzzy
representation learning with tsk fuzzy system,” IEEE Transactions on Systems, vol. 28, no. 10, pp. 2348–2362, 2020.
Fuzzy Systems, 2019. 211 Garcı́a-Vico, Á. M., Carmona, C. J., Gonzalez, P., Seker, H., and del
186 Zuo, H., Zhang, G., Pedrycz, W., Behbood, V., and Lu, J., “Fuzzy Jesus, M. J., “Fepds: A proposal for the extraction of fuzzy emerging
regression transfer learning in takagi–sugeno fuzzy models,” IEEE patterns in data streams,” IEEE Transactions on Fuzzy Systems, vol. 28,
Transactions on Fuzzy Systems, vol. 25, no. 6, pp. 1795–1807, 2017. no. 12, pp. 3193–3203, 2020.
187 Zuo, H., Zhang, G., Lu, J., and Pedrycz, W., “Fuzzy rule-based transfer 212 Lughofer, E. and Angelov, P., “Handling drifts and shifts in on-line data
learning for label space adaptation,” in 2017 IEEE International Confer- streams with evolving fuzzy systems,” Applied Soft Computing, vol. 11,
ence on Fuzzy Systems, 2017, pp. 1–6. no. 2, pp. 2057–2068, 2011.
213 Yu, H., Lu, J., and Zhang, G., “Topology learning-based fuzzy random
188 Zuo, H., Zhang, G., Pedrycz, W., Behbood, V., and Lu, J., “Granular
neural networks for streaming data regression,” IEEE Transactions on
fuzzy regression domain adaptation in takagi–sugeno fuzzy models,”
Fuzzy Systems, vol. 30, no. 2, pp. 412–425, 2020.
IEEE Transactions on Fuzzy Systems, vol. 26, no. 2, pp. 847–858, 2018.
214 Ge, D. and Zeng, X.-J., “Learning data streams online—an evolving
189 Zuo, H., Lu, J., Zhang, G., and Liu, F., “Fuzzy transfer learning using an
fuzzy system approach with self-learning/adaptive thresholds,” Informa-
infinite gaussian mixture model and active learning,” IEEE Transactions
tion sciences, vol. 507, pp. 172–184, 2020.
on Fuzzy Systems, vol. 27, no. 2, pp. 291–303, 2019.
215 Gu, X. and Shen, Q., “A self-adaptive fuzzy learning system for
190 Li, K., Lu, J., Zuo, H., and Zhang, G., “Multi-source domain adaptation
streaming data prediction,” Information Sciences, vol. 579, pp. 623–647,
with fuzzy-rule based deep neural networks,” in 2021 IEEE International
2021.
Conference on Fuzzy Systems, 2021, pp. 1–6.
216 Gu, X., “An explainable semi-supervised self-organizing fuzzy inference
191 Li, K., Lu, J., Zuo, H., and Zhang, G., “Source-free multi-domain adap- system for streaming data classification,” Information Sciences, vol. 583,
tation with fuzzy rule-based deep neural networks,” IEEE Transactions pp. 364–385, 2022.
on Fuzzy Systems, 2023. 217 Ferdaus, M. M., Pratama, M., Anavatti, S. G., and Garratt, M. A., “Palm:
192 Zuo, H., Lu, J., Zhang, G., and Pedrycz, W., “Fuzzy rule-based domain An incremental construction of hyperplanes for data stream regression,”
adaptation in homogeneous and heterogeneous spaces,” IEEE Transac- IEEE Transactions on Fuzzy Systems, vol. 27, no. 11, pp. 2115–2129,
tions on Fuzzy Systems, vol. 27, no. 2, pp. 348–361, 2019. 2019.
193 Che, X., Zuo, H., Lu, J., and Chen, D., “Fuzzy multioutput transfer 218 Leite, D., Andonovski, G., Škrjanc, I., and Gomide, F., “Optimal rule-
learning for regression,” IEEE Transactions on Fuzzy Systems, vol. 30, based granular systems from data streams,” IEEE Transactions on Fuzzy
no. 7, pp. 2438–2451, 2021. Systems, vol. 28, no. 3, pp. 583–596, 2019.
194 Pal, S. K. and Ghosh, A., “Fuzzy geometry in image analysis,” Fuzzy 219 Liu, Y., Zhao, J., Wang, W., and Pedrycz, W., “Prediction intervals for
Sets and Systems, vol. 48, no. 1, pp. 23–40, 1992. granular data streams based on evolving type-2 fuzzy granular neural
195 Yang, J., Wang, G., and Zhang, Q., “Knowledge distance measure in network dynamic ensemble,” IEEE Transactions on Fuzzy Systems,
multigranulation spaces of fuzzy equivalence relations,” Information vol. 29, no. 4, pp. 874–888, 2020.
Sciences, vol. 448, pp. 18–35, 2018. 220 e Silva, P. C. d. L., Junior, C. A. S., Alves, M. A., Silva, R.,
196 Liu, F., Zhang, G., and Lu, J., “A novel fuzzy neural network for Cohen, M. W., and Guimarães, F. G., “Forecasting in non-stationary
unsupervised domain adaptation in heterogeneous scenarios,” in 2019 environments with fuzzy time series,” Applied Soft Computing, vol. 97,
IEEE International Conference on Fuzzy Systems, 2019, pp. 1–6. p. 106825, 2020.
197 Ma, G., Lu, J., and Zhang, G., “Multi-source domain adaptation with 221 Severiano, C. A., e Silva, P. C. d. L., Cohen, M. W., and Guimarães, F. G.,
interval-valued target data via fuzzy neural networks,” IEEE Transactions “Evolving fuzzy time series for spatio-temporal forecasting in renewable
on Fuzzy Systems, 2024. energy systems,” Renewable Energy, vol. 171, pp. 764–783, 2021.
198 Li, K., Lu, J., Zuo, H., and Zhang, G., “Attention-bridging ts fuzzy rules 222 Liu, A., Lu, J., and Zhang, G., “Concept drift detection: dealing with
for universal multi-domain adaptation without source data,” in 2023 IEEE missing values via fuzzy distance estimations,” IEEE Transactions on
International Conference on Fuzzy Systems, 2023, pp. 1–6. Fuzzy Systems, vol. 29, no. 11, pp. 3219–3233, 2020.
199 Ma, G., Lu, J., Liu, F., Fang, Z., and Zhang, G., “Domain adaptation with 223 Liu, A., Zhang, G., and Lu, J., “Fuzzy time windowing for gradual
interval-valued observations: Theory and algorithms,” IEEE Transactions concept drift adaptation,” in 2017 IEEE International Conference on
on Fuzzy Systems, 2024. Fuzzy Systems, 2017, pp. 1–6.
200 Ma, G., Lu, J., and Zhang, G., “Interval-valued observations-based multi- 224 Song, Y., Zhang, G., Lu, H., and Lu, J., “A fuzzy drift correlation
source domain adaptation using fuzzy neural networks,” in 2023 IEEE matrix for multiple data stream regression,” in 2020 IEEE International
International Conference on Fuzzy Systems, 2023. Conference on Fuzzy Systems, 2020, pp. 1–6.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

17

225 Dong, F., Lu, J., Zhang, G., and Li, K., “Active fuzzy weighting ensemble 249 Yera, R., Alzahrani, A. A., and Martı́nez, L., “A fuzzy content-based
for dealing with concept drift,” International Journal of Computational group recommender system with dynamic selection of the aggregation
Intelligence Systems, vol. 11, no. 1, pp. 438–450, 2018. functions,” International Journal of Approximate Reasoning, vol. 150,
226 Zhang, N., Yao, X., and Luo, C., “The prediction of online time series pp. 273–296, 2022.
with concept drift based on dynamic intuitionistic fuzzy cognitive map,” 250 Sulthana, A. R. and Ramasamy, S., “Ontology and context based
Intelligent Data Analysis, vol. 25, no. 4, pp. 949–972, 2021. recommendation system using neuro-fuzzy classification,” Computers &
227 Yu, H., Lu, J., Liu, A., Wang, B., Li, R., and Zhang, G., “Real-time Electrical Engineering, vol. 74, pp. 498–510, 2019.
prediction system of train carriage load based on multi-stream fuzzy 251 Patro, S. G. K., Mishra, B. K., Panda, S. K., Kumar, R., Long, H. V.,
learning,” IEEE Transactions on Intelligent Transportation Systems, and Tuan, T. M., “Knowledge-based preference learning model for
vol. 23, no. 9, pp. 15 155–15 165, 2022. recommender system using adaptive neuro-fuzzy inference system,”
228 Song, Q. and Chissom, B. S., “Fuzzy time series and its models,” Fuzzy Journal of Intelligent & Fuzzy Systems, vol. 39, no. 3, pp. 4651–4665,
sets and systems, vol. 54, no. 3, pp. 269–277, 1993. 2020.
229 Lütjens, B., Everett, M., and How, J. P., “Safe reinforcement learning 252 Zhang, W., Zhang, X., and Chen, D., “Causal neural fuzzy infer-
with model uncertainty estimates,” in 2019 International Conference on ence modeling of missing data in implicit recommendation system,”
Robotics and Automation, 2019, pp. 8662–8668. Knowledge-Based Systems, vol. 222, p. 106678, 2021.
230 Devidze, R., Kamalaruban, P., and Singla, A., “Exploration-guided re- 253 Walek, B. and Fajmon, P., “A hybrid recommender system for an online
ward shaping for reinforcement learning under sparse rewards,” Advances store using a fuzzy expert system,” Expert Systems with Applications,
in Neural Information Processing Systems, vol. 35, pp. 5829–5842, 2022. vol. 212, p. 118565, 2023.
254 Kuanr, M., Rath, B. K., and Mohanty, S. N., “Crop recommender system
231 Zhang, H., Zhang, K., Cai, Y., and Han, J., “Adaptive fuzzy fault-tolerant
for the farmers using mamdani fuzzy inference model,” International
tracking control for partially unknown systems with actuator faults via
Journal of Engineering & Technology, vol. 7, no. 4.15, pp. 277–280,
integral reinforcement learning method,” IEEE Transactions on Fuzzy
2018.
Systems, vol. 27, no. 10, pp. 1986–1998, 2019.
255 Castro, J., Yera, R., and Martinez, L., “A fuzzy approach for natural
232 Zhang, K., Zhang, H., Mu, Y., and Liu, C., “Decentralized tracking opti- noise management in group recommender systems,” Expert Systems with
mization control for partially unknown fuzzy interconnected systems via Applications, vol. 94, pp. 237–249, 2018.
reinforcement learning method,” IEEE Transactions on Fuzzy Systems, 256 Rutkowski, T., Romanowski, J., Woldan, P., Staszewski, P., Nielek, R.,
vol. 29, no. 4, pp. 917–926, 2020. and Rutkowski, L., “A content-based recommendation system using
233 Kumar, N., Rahman, S. S., and Dhakad, N., “Fuzzy inference enabled neuro-fuzzy approach,” in 2018 IEEE International Conference on Fuzzy
deep reinforcement learning-based traffic light control for intelligent Systems, 2018, pp. 1–8.
transportation system,” IEEE Transactions on Intelligent Transportation 257 Almohammadi, K., Hagras, H., Yao, B., Alzahrani, A., Alghazzawi, D.,
Systems, vol. 22, no. 8, pp. 4919–4928, 2020. and Aldabbagh, G., “A type-2 fuzzy logic recommendation system for
234 Li, J., Shi, H., and Hwang, K., “Using fuzzy logic to learn abstract poli- adaptive teaching,” Soft Computing, vol. 21, pp. 965–979, 2017.
cies in large-scale multiagent reinforcement learning,” IEEE Transactions 258 Nagaraj, P. and Deepalakshmi, P., “An intelligent fuzzy inference rule-
on Fuzzy Systems, vol. 30, no. 12, pp. 5211–5224, 2022. based expert recommendation system for predictive diabetes diagnosis,”
235 Zhu, A., Ai, H., and Chen, L., “A fuzzy logic reinforcement learning International Journal of Imaging Systems and Technology, vol. 32, no. 4,
control with spring-damper device for space robot capturing satellite,” pp. 1373–1396, 2022.
Applied Sciences, vol. 12, no. 5, p. 2662, 2022. 259 Karthik, R. and Ganapathy, S., “A fuzzy recommendation system for
236 Fang, B., Zheng, C., Wang, H., and Yu, T., “Two stream fused fuzzy predicting the customers interests using sentiment analysis and ontology
deep neural network for multi-agent learning,” IEEE Transactions on in e-commerce,” Applied Soft Computing, vol. 108, p. 107396, 2021.
Fuzzy Systems, 2022. 260 Yan, S., Pirooznia, S., Heidari, A., Navimipour, N. J., and Unal, M.,
237 Fathinezhad, F., Derhami, V., and Rezaeian, M., “Supervised fuzzy “Implementation of a product-recommender system in an iot-based smart
reinforcement learning for robot navigation,” Applied Soft Computing, shopping using fuzzy logic and apriori algorithm,” IEEE Transactions
vol. 40, pp. 33–41, 2016. on Engineering Management, 2022.
238 Hein, D., Hentschel, A., Runkler, T., and Udluft, S., “Particle swarm 261 Kermany, N. R. and Alizadeh, S. H., “A hybrid multi-criteria recom-
optimization for generating interpretable fuzzy reinforcement learning mender system using ontology and neuro-fuzzy techniques,” Electronic
policies,” Engineering Applications of Artificial Intelligence, vol. 65, pp. Commerce Research and Applications, vol. 21, pp. 50–64, 2017.
87–98, 2017. 262 Manogaran, G., Varatharajan, R., and Priyan, M. K., “Hybrid recom-
239 Yin, L. and Li, Y., “Fuzzy vector reinforcement learning algorithm mendation system for heart disease diagnosis based on multiple kernel
for generation control of power systems considering flywheel energy learning with adaptive neuro-fuzzy inference system,” Multimedia tools
storage,” Applied Soft Computing, vol. 125, p. 109149, 2022. and applications, vol. 77, pp. 4379–4399, 2018.
240 Shi, H., Lin, Z., Zhang, S., Li, X., and Hwang, K., “An adaptive decision- 263 Thong, N. T. et al., “Intuitionistic fuzzy recommender systems: an
making method with fuzzy bayesian reinforcement learning for robot effective tool for medical diagnosis,” Knowledge-Based Systems, vol. 74,
soccer,” Information Sciences, vol. 436, pp. 268–281, 2018. pp. 133–150, 2015.
241 Atanassov, K., “Intuitionistic fuzzy sets,” Fuzzy Sets and Systems, 264 Guo, J., Deng, J., and Wang, Y., “An intuitionistic fuzzy set based
vol. 20, pp. 87–96, 1986. hybrid similarity model for recommender system,” Expert Systems with
Applications, vol. 135, pp. 153–163, 2019.
242 Lu, J., Zhu, Y., Zeng, X., Koehl, L., Ma, J., and Zhang, G., “A linguistic
265 Yera, R., Castro, J., and Martı́nez, L., “A fuzzy model for managing
multi-criteria group decision support system for fabric hand evaluation,”
natural noise in recommender systems,” Applied Soft Computing, vol. 40,
Fuzzy Optimization and Decision Making, vol. 8, pp. 395–413, 2009.
pp. 187–198, 2016.
243 Ishibuchi, H. and Nakashima, T., “Effect of rule weights in fuzzy rule- 266 Wang, P., Wang, Y., Zhang, L. Y., and Zhu, H., “An effective and efficient
based classification systems,” IEEE Transactions on Fuzzy Systems, fuzzy approach for managing natural noise in recommender systems,”
vol. 9, no. 4, pp. 506–515, 2001. Information Sciences, vol. 570, pp. 623–637, 2021.
244 Cornelis, C., Lu, J., Guo, X., and Zhang, G., “One-and-only item 267 Saravanan, B., Mohanraj, V., and Senthilkumar, J., “A fuzzy entropy
recommendation with fuzzy logic techniques,” Information sciences, vol. technique for dimensionality reduction in recommender systems using
177, no. 22, pp. 4906–4921, 2007. deep learning,” Soft Computing, vol. 23, pp. 2575–2583, 2019.
245 Katarya, R. and Verma, O. P., “An effective web page recommender sys- 268 Zenebe, A. and Norcio, A. F., “Representation, similarity measures and
tem with fuzzy c-mean clustering,” Multimedia Tools and Applications, aggregation methods using fuzzy sets for content-based recommender
vol. 76, pp. 21 481–21 496, 2017. systems,” Fuzzy sets and systems, vol. 160, no. 1, pp. 76–94, 2009.
246 Katarya, R. and Verma, O., “Recommender system with grey wolf 269 Shang, F., Liu, Y., Cheng, J., and Yan, D., “Fuzzy double trace norm
optimizer and fcm,” Neural Computing and Applications, vol. 30, pp. minimization for recommendation systems,” IEEE Transactions on Fuzzy
1679–1687, 2018. Systems, vol. 26, no. 4, pp. 2039–2049, 2017.
247 Selvi, C. and Sivasankar, E., “A novel optimization algorithm for rec- 270 Mezei, J. and Nikou, S., “Fuzzy optimization to improve mobile health
ommender system using modified fuzzy c-means clustering approach,” and wellness recommendation systems,” Knowledge-Based Systems, vol.
Soft Computing, vol. 23, pp. 1901–1916, 2019. 142, pp. 108–116, 2018.
248 Nilashi, M., Ibrahim, O. B., Ithnin, N., and Zakaria, R., “A multi-criteria 271 Huitzil, I., Alegre, F., and Bobillo, F., “Gimmehop: A recommender
recommendation system using dimensionality reduction and neuro-fuzzy system for mobile devices using ontology reasoners and fuzzy logic,”
techniques,” Soft Computing, vol. 19, pp. 3173–3207, 2015. Fuzzy Sets and Systems, vol. 401, pp. 55–77, 2020.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Transactions on Fuzzy Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TFUZZ.2024.3387429

18

272 Abbasi-Moud, Z., Hosseinabadi, S., Kelarestaghi, M., and Eshghi, F., Guangquan Zhang is an Australian Research Coun-
“Cafob: Context-aware fuzzy-ontology-based tourism recommendation cil (ARC) QEII Fellow, Associate Professor and
system,” Expert Systems with Applications, vol. 199, p. 116877, 2022. the Director of the Decision Systems and e-Service
273 Zhang, Z., Lin, H., Liu, K., Wu, D., Zhang, G., and Lu, J., “A Intelligent (DeSI) Research Laboratory at the Aus-
hybrid fuzzy-based personalized recommender system for telecom prod- tralian Artificial Intelligence Institute, University of
ucts/services,” Information Sciences, vol. 235, pp. 117–129, 2013. Technology Sydney, Australia. He received his PhD
274 Wu, D., Lu, J., and Zhang, G., “A fuzzy tree matching-based personalized in applied mathematics from Curtin University, Aus-
e-learning recommender system,” IEEE transactions on fuzzy systems, tralia, in 2001. From 1993 to 1997, he was a full
vol. 23, no. 6, pp. 2412–2426, 2015. Professor in the Department of Mathematics, Hebei
275 Martinez-Cruz, C., Porcel, C., Bernabé-Moreno, J., and Herrera- University, China. His main research interests lie
Viedma, E., “A model to represent users trust in recommender systems in fuzzy multi-objective, bilevel and group decision
using ontologies and fuzzy linguistic modeling,” Information Sciences, making, fuzzy measures, transfer learning and concept drift adaptation. He has
vol. 311, pp. 102–118, 2015. published six authored monographs and over 500 papers including some 300
276 Gao, R., Wang, J., Zhou, K., Liu, F., Xie, B., Niu, G., Han, B., articles in leading international journals. He has supervised 40 PhD students
and Cheng, J., “Fast and reliable evaluation of adversarial robustness to completion and mentored 15 Postdoc fellows. Prof Zhang has won ten
with minimum-margin attack,” in International Conference on Machine very competitive ARC Discovery grants and many other research projects.
Learning, 2022, pp. 7144–7163. His research has been widely applied in industries.
277 Zuo, H., Zhang, G., and Lu, J., “Fuzzy transfer learning in heterogeneous
space using takagi-sugeno fuzzy models,” in International Fuzzy Systems
Association World Congress, 2019, pp. 752–763.
278 Fang, Z., Lu, J., Liu, A., Liu, F., and Zhang, G., “Learning bounds
for open-set learning,” in International conference on machine learning,
2021, pp. 3122–3132.
279 Zang, T., Zhu, Y., Liu, H., Zhang, R., and Yu, J., “A survey on cross-
domain recommendation: taxonomies, methods, and future directions,”
ACM Transactions on Information Systems, vol. 41, no. 2, pp. 1–39,
2022.

Jie Lu (F’18) is an Australian Laureate Fellow,

IFSA Fellow, Distinguished Professor and the Di-
rector of Australian Artificial Intelligence Institute
(AAII) at the University of Technology Sydney,
Australia. She received a PhD degree from Curtin
University in 2000. Her main research expertise is in
transfer learning, concept drift, fuzzy systems, deci-
sion support systems and recommender systems. She
has published over 500 papers in IEEE Transactions
and other leading journals and conferences. She is
the recipient of two IEEE Transactions on Fuzzy
Systems Outstanding Paper Awards (2019 and 2022), NeurIPS2022 Outstand-
ing Paper Award, Australia’s Most Innovative Engineer Award (2019), Aus-
tralasian Artificial Intelligence Distinguished Research Contribution Award
(2022), Australian NSW Premier’s Prize on Excellence in Engineering or
Information & Communication Technology (2023), and the Officer of the
Order of Australia (AO) 2023.

Guangzhi Ma received his M.S. degree in probabil-

ity and statistics from the School of Mathematics and
Statistics, Lanzhou University, Lanzhou, China, in
2020. He is now a PhD candidate with the Faculty of
Engineering and Information Technology, University
of Technology Sydney, Australia. He is a Member
of the Decision Systems and e-Service Intelligence
Lab, Australia Artificial Intelligence Institute, Uni-
versity of Technology Sydney. His research interests
include fuzzy transfer learning, domain adaptation
and interval-valued data analysis. He has published
several paper related to interval-valued data and transfer learning in FUZZ-
IEEE, TFS, TCYB.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]