Journal of Computer and Communications, 2024, 12, 53-75

[Link]
ISSN Online: 2327-5227
ISSN Print: 2327-5219

Missing Data Imputation: A Comprehensive

Review

Majed Alwateer1, El-Sayed Atlam1,2, Mahmoud Mohammed Abd El-Raouf3, Osama A. Ghoneim4,
Ibrahim Gad2
1
Department of Computer Science, College of Computer Science and Engineering, Taibah University, Yanbu, Saudi Arabia
2
Computer Science Department, Faculty of Science, Tanta University, Tanta, Egypt
3
Basic and Applied Science Institute, College of Engineering and Technology, Arab Academy for Science and Technology (AAST),
Alexandria, Egypt
4
Department of Computer Science, Faculty of Computers and informatics, Tanta University, Tanta, Egypt

How to cite this paper: Alwateer, M., At- Abstract

lam, E.-S., Abd El-Raouf, M.M., Ghoneim,
O.A. and Gad, I. (2024) Missing Data Impu- Missing data presents a significant challenge in statistical analysis and ma-
tation: A Comprehensive Review. Journal of chine learning, often resulting in biased outcomes and diminished efficiency.
Computer and Communications, 12, 53-75.
This comprehensive review investigates various imputation techniques, cate-
[Link]
gorizing them into three primary approaches: deterministic methods, proba-
Received: September 25, 2024 bilistic models, and machine learning algorithms. Traditional techniques, in-
Accepted: November 8, 2024 cluding mean or mode imputation, regression imputation, and last observa-
Published: November 11, 2024
tion carried forward, are evaluated alongside more contemporary methods
Copyright © 2024 by author(s) and such as multiple imputation, expectation-maximization, and deep learning
Scientific Research Publishing Inc. strategies. The strengths and limitations of each approach are outlined. Key
This work is licensed under the Creative
considerations for selecting appropriate methods, based on data characteris-
Commons Attribution International
License (CC BY 4.0). tics and research objectives, are discussed. The importance of evaluating im-
[Link] putation’s impact on subsequent analyses is emphasized. This synthesis of re-
Open Access cent advancements and best practices provides researchers with a robust
framework for effectively handling missing data, thereby improving the relia-
bility of empirical findings across diverse disciplines.

Keywords
Missing Data, Machine Learning, Prediction, Deep Learning, Imputation

1. Introduction
Missing data can occur due to various reasons, such as participant non-response
in surveys, equipment malfunction in experimental settings, or data entry errors

DOI: 10.4236/jcc.2024.1211004 Nov. 11, 2024 53 Journal of Computer and Communications

M. Alwateer et al.

[1] [2]. The presence of missing data can significantly impact statistical analyses
and lead to incorrect conclusions if not properly addressed [3]. Missing data is a
pervasive issue in empirical research across various disciplines, including social
sciences, medical research, and data science [4]. It occurs when no value is avail-
able for a variable in an observation, potentially leading to incomplete datasets
that can compromise the validity and reliability of statistical analyses [5] [6].
Missing values can significantly impact your analysis: They can introduce bias
if not handled properly. Many machine learning algorithms can’t handle missing
values that are out of the box. They can lead to the loss of important information
if instances with missing values are simply discarded. Improperly handled missing
values can lead to incorrect conclusions or predictions.
Missing values can sneak into your data for a variety of reasons. Here are some
common reasons: Data Entry Errors: Sometimes, it’s just human error. Someone
might forget to input a value or accidentally delete one. Sensor Malfunctions: In
IoT or scientific experiments, a faulty sensor might fail to record data at certain
times [7]. Survey Non-Response: In surveys, respondents might skip questions
they are uncomfortable answering or don’t understand. Merged Datasets: When
combining data from multiple sources, some entries might not have correspond-
ing values in all datasets. Data Corruption: During data transfer or storage, some
values might get corrupted and become unreadable. Intentional Omissions: Some
data might be intentionally left out due to privacy concerns or irrelevance. Sam-
pling Issues: The data collection method might systematically miss certain types
of data. Time-Sensitive Data: In time series data, values might be missing for pe-
riods when data wasn’t collected (e.g., weekends, holidays) [8].
Researchers typically encounter three main types of missing data mechanisms, as
defined by Rubin [4]: 1) Missing Completely at Random (MCAR): The probability
of missing data is unrelated to both observed and unobserved variables. 2) Missing
at Random (MAR): The probability of missing data depends on observed variables
but not on unobserved variables. 3) Missing Not at Random (MNAR): The probabil-
ity of missing data depends on unobserved variables, including the missing data itself.
Addressing missing data is of paramount importance in research for several
reasons: 1) Bias reduction: Ignoring missing data or using simplistic methods like
complete case analysis can lead to biased estimates and incorrect inferences [9].
Proper handling of missing data helps minimize this bias and improve the accu-
racy of research findings. 2) Statistical power: Missing data reduces the effective
sample size, leading to decreased statistical power. Appropriate imputation tech-
niques can help maintain or even increase power by utilizing all available infor-
mation [10]. 3) Generalizability: Incomplete datasets may not be representative of
the population of interest, potentially limiting the generalizability of research
findings. Addressing missing data can help improve the external validity of results
[11]. 4) Ethical considerations: In clinical trials and other human subjects re-
search, ignoring missing data may lead to the waste of valuable data that partici-
pants have provided, raising ethical concerns. 5) Regulatory compliance: In some
fields, such as clinical trials, regulatory bodies require proper handling and reporting

DOI: 10.4236/jcc.2024.1211004 54 Journal of Computer and Communications

 M. Alwateer et al.

of missing data [12]. 6) Improved decision-making: In applied settings, such as

business analytics or public policy, accurate and complete data is essential for in-
formed decision-making [13].
While imputation methods offer valuable tools for handling missing data, sev-
eral challenges and considerations must be addressed to ensure effective and reli-
able results: 1) Handling different data types: Imputation methods must be able
to handle various data types, including continuous, categorical, and mixed data
[14]. 2) Dealing with high-dimensional data: As the number of variables increases,
imputation becomes more challenging due to the curse of dimensionality [15]. 3)
Computational efficiency: Some advanced imputation methods can be computa-
tionally intensive, necessitating efficient implementations for large-scale applica-
tions [16]. 4) Preserving data distributions and relationships: Imputation methods
should maintain the statistical properties of the original data, including distribu-
tions and relationships between variables [17].
Missing data is a pervasive issue in research across various disciplines, often
leading to biased or inefficient analyses [4]. Addressing missing data is crucial for
maintaining the integrity and reliability of research findings [18]. This review
aims to provide a comprehensive overview of missing data imputation techniques,
their applications, and current challenges in the field.
Table 1 summarizing the recent papers on imputation methods published be-
tween 2017 and 2024, focusing on the year, model used, columns imputed, and
key results.
Given the critical nature of missing data in research, this comprehensive review
aims to achieve the following objectives: 1) Provide an up-to-date synthesis of
current missing data imputation techniques, including traditional methods and
advanced machine learning approaches. 2) Critically evaluate the strengths and
limitations of various imputation methods across different research contexts and
data types. 3) Examine the impact of different missing data mechanisms on the
performance of imputation techniques. 4) Explore emerging trends and future di-
rections in missing data imputation, including the application of deep learning
and artificial intelligence techniques.
Offer practical guidelines to help researchers select appropriate imputation
methods based on their specific research context, data characteristics, and analysis
goals. Identify gaps in the current literature and propose areas for future research
in missing data imputation.
 Provide an overview of traditional and advanced imputation methods.
 Discuss evaluation metrics for imputation techniques.
 Examine challenges and considerations in missing data imputation.
 Explore case studies and applications across various domains.
 Identify future directions and open problems in the field data imputation.
By addressing these objectives, this review aims to provide researchers, statisti-
cians, and data scientists with a comprehensive understanding of missing data
imputation techniques, their applications, and their implications for research in-
tegrity and validity.

DOI: 10.4236/jcc.2024.1211004 55 Journal of Computer and Communications

M. Alwateer et al.

Table 1. Summary of imputation methods in published papers (2017-2024).

Paper title Year Model used Columns imputed Key results

Accounted for temporal dependencies in

kernel ridge, linear Multiple variables
Doreswamy et al. longitudinal data, leading to more accurate
2017 regression, random of NCDC weather
[19] estimates of parameters and improved
forest, SVM, and KNN dataset
model performance.

Improved accuracy of predictive models

Machine learning NCDC weather
Hosahalli et al. [20] 2018 for NCDC weather dataset compared to
models datasets
single imputation methods.

Demonstrated effectiveness in imputing

Predictive mean missing values in educational data from a
Khanani [21] 2021 Education data
matching (PMM) public school with both numerical and
categorical features.

Provided a comprehensive overview of

multiple imputation techniques and their
Thakur et al. [22] 2021 Machine learning Time series data
applications in machine learning, highlighting
their advantages and limitations.

Improved accuracy of predictive models

Psychogyios et al. Age, gender, diagnosis
2023 KNN-MICE-GAN for hospital readmission compared to
[23] codes, lab results
single imputation methods.

Random forest, The results showed that the Random Forest

decision tree, neural net- Dropout in higher algorithm obtained the best performance,
Omar et al. [24] 2023
work and support vector education with an AUC of 0.9623 in the prediction
machine of college dropout.

The KNN achieved high imputation

Mean imputation, accuracy for missing rainfall variables
Nida et al. [25] 2023 Rainfall data
KNN, PMM values, improving the analysis of complex
weather datasets.

The results show that GAN achieved high

Psychogyios et al. Electronic health
2023 GAN imputation accuracy and outperform the
[23] records
standard baselines.

Provided a comprehensive overview of

multiple imputation techniques for
Teegavarapu et al. Spatial and temporal Hydrometeorological
2024 precipitation, temperature, and streamflows
[26] interpolation methods data
and highlighting their advantages and
limitations.

The results demonstrated that the

Focalize K-NN
Almeida et al. [27] 2024 Time series data effectiveness of Focalize K-NN for imputing
method
missing values in time series data.

The proposed imputation method

demonstrates superior performance across
Self-attention Electronic health
Kowsar et al. [28] 2024 a range of missing data proportions (10%
imputation method records
to 50%) under the assumption of missing
completely at random (MCAR).

DOI: 10.4236/jcc.2024.1211004 56 Journal of Computer and Communications

 M. Alwateer et al.

2. Missing Data Types

Understanding the mechanisms of missingness is essential for selecting appropri-
ate imputation methods and accurately interpreting results. Rubin’s classification
system for missing data, which remains fundamental to the field, identifies three
main types of missing data [18].

2.1. Missing Completely at Random (MCAR)

Missingness is considered completely random when it does not depend on any
other variables. This condition, known as Missing Completely at Random
(MCAR), occurs when the probability of missing data is unrelated to both ob-
served and unobserved variables [29]-[31]. In this scenario, the missingness is
purely due to chance, with no influence from any characteristics of the data [4].
MCAR is the most stringent assumption and rarely occurs in practice. However,
when data are MCAR, analyses using only complete cases will be unbiased, alt-
hough potentially inefficient [1].
Let R be the missingness indicator and Y be the complete data. MCAR can
be defined as: P ( R | Y ) = P ( R ) .
For example, in a survey, some participants accidentally skip questions regard-
less of their characteristics or responses to other items.

2.2. Missing at Random (MAR)

The probability of missing data depends on other observed variables but not on
the missing data itself. This condition, known as Missing at Random (MAR), oc-
curs when the likelihood of missingness is related to observed variables but not to
unobserved ones [18]. MAR represents a less stringent assumption compared to
Missing Completely at Random (MCAR) and is often more realistic in practice.
In summary, data are considered MAR when the probability of missingness is in-
fluenced by observed variables while remaining independent of unobserved vari-
ables [18]. MAR allows for relationships between observed variables and the prob-
ability of missingness. Many modern imputation methods, such as multiple im-
putation, assume MAR [17].
Let Yobs be the observed data and Ymis be the missing data. MAR can be de-
fined as: P ( R | Y ) = P ( R | Yobs ) .
For example, in a longitudinal study on income, older participants are more
likely to withhold information about their earnings, but this likelihood is not re-
lated to the actual income amount after accounting for age. For example, men
might be less likely to answer questions about emotions in a survey.

2.3. Missing Not at Random (MNAR)

Missingness is categorized as Missing Not at Random (MNAR) when the proba-
bility of missing data is influenced by the values of the missing data itself or by un-
observed variables. In this case, the missingness is directly related to the unobserved

DOI: 10.4236/jcc.2024.1211004 57 Journal of Computer and Communications

M. Alwateer et al.

values [4]. MNAR is the most challenging type of missing data to handle. Standard
imputation methods can yield biased results, necessitating specialized techniques
or sensitivity analyses to address the issue effectively [32].
MNAR occurs when the probability of missing data depends on unobserved
variables or the missing values themselves [4]. This is the most challenging type
of missingness to address and often requires specialized techniques. MNAR oc-
curs when: P ( R | Y ) ≠ P ( R | Yobs ) .
An example of MNAR is in a mental health survey, participants with severe
depression may be less likely to complete questions about their symptoms, with
this likelihood directly related to the severity of their undisclosed condition. Sim-
ilarly, individuals with high incomes might be less inclined to report their income
in a survey.
In practice, it is often impossible to definitively determine whether data are
MAR or MNAR based solely on observed data [33]. Therefore, researchers often
must rely on subject-matter knowledge and conduct sensitivity analyses to assess
the robustness of their findings under different missing data assumptions [17].

3. Missing Value Imputation

Figure 1. The flowchart of imputation of missing values.

The problem of missing data imputation arises when a dataset contains unob-
=
served values for certain features. Let X ( X 1 , X 2 , , X n ) ∈  n represent a ran-
dom vector, where each X i corresponds to a feature and follows a distribution
P ( X ) . A binary mask vector M = ( M 1 , M 2 , , M n ) indicates the presence or

DOI: 10.4236/jcc.2024.1211004 58 Journal of Computer and Communications

 M. Alwateer et al.

absence of observations, with M i = 1 denoting an observed value and M i = 0 a

missing value. Given a dataset of d instances {( X i , M i=
)}i 0,,d −1 , an imputed da-
taset is constructed by replacing missing values (where M ij = 0 ) in the observed
data X i with pre-imputed values, potentially random noise, resulting in Xˆ i .
The objective is to develop an imputation model, IMP, that generates an imputed
{ (
dataset X i = IMP Xˆ i , M i )}
=i 0,,d −1
such that each imputed sample X i is
drawn from the conditional distribution P( X | X ) , thereby preserving the orig-
i

inal data’s distributional properties. This yields a complete dataset X , where

=X X  M + (1 − M )  X for each sample i .
i i i i i

Following imputation of missing values, predictive modeling was performed as

shown in Figure 1. Application of S distinct imputation algorithms,
A0 , A1 , , AS −1 , to the original dataset X yielded S completed datasets, de-
noted D1 , D2 , , DS . A standard predictive model, P , was then applied to each
of these completed datasets to predict the outcome.

4. Methods
This study evaluated several imputation methods for handling missing data, rang-
ing from simple statistical techniques to more sophisticated deep learning ap-
proaches. Specific methods included.

4.1. Traditional Imputation Methods

Traditional methods for handling missing data have been widely used due to their
simplicity and ease of implementation [1]. Traditional imputation methods are
widely used to handle missing data in datasets. These methods aim to replace
missing values with plausible estimates based on the available data. The most com-
mon traditional imputation techniques are discussed as follows.

4.1.1. Mean/Median Imputation

Missing categorical values were replaced with the mode, while missing numerical
values were replaced with the mean of the corresponding feature. This method
involves replacing missing values with the mean or median of the observed values
for that variable [4]. Although simple, this approach can result in biased estimates
and an underestimation of standard errors. For numerical data, missing values are
replaced with the mean or median of the non-missing values in the same column,
while for categorical data, the mode (most frequent value) is utilized.

4.1.2. MissForest
This iterative approach uses mean/mode imputation to initialize the dataset [16].
Then, a random forest is trained to predict the missing values in each feature,
iteratively refining the imputed values until convergence (defined by a lack of im-
provement in the imputed matrix). Convergence criteria included a maximum of
20 iterations and 100 trees. The difference between successive imputed matrices
( M imp _ new and M imp _ old ) for numerical ( N ) and categorical ( F ) features was
measured as:

DOI: 10.4236/jcc.2024.1211004 59 Journal of Computer and Communications

M. Alwateer et al.

∑ ( M new )
imp imp 2
− M old
δ N = j∈N (1)
∑ j∈N ( M new )
imp 2

i =n
∑ j∈F ‍ ∑ i=I‍
1 M imp
≠M imp
δF = new old
(2)
FNA

where FNA represents the number of missing values in categorical variables.

4.1.3. LOCF and NOCB

Last Observation Carried Forward (LOCF) and Next Observation Carried Back-
ward (NOCB) are both methods used for handling missing data in longitudinal
studies. LOCF fills in missing data points by carrying forward the last observed
value. For example, if a participant’s value is missing at a follow-up, the last rec-
orded value is used to fill in that gap [34].
The advantages of LOCF method are: 1) Simplicity: Easy to implement and un-
derstand. 2) Preservation of Sample Size: Retains all participants in the analysis,
which can be important in clinical trials. On the other hand, the disadvantages of
LOCF method are: 1) Assumption of Stability: Implies that the last observation is
a good estimate for future values, which may not hold true. 2) Potential Bias: Can
introduce bias if the last observed value is not representative of the participant’s
state at the time of the missing data. 3) Underestimation of Variability: Fails to
account for natural fluctuations in the data, potentially leading to misleading con-
clusions.
NOCB fills in missing data points by carrying the next observed value back-
ward. For example, if a participant’s value is missing before a subsequent obser-
vation, the next recorded value is used to fill in the gap.
The advantages of NOCB method are: 1) Preservation of Trends: Can better
reflect changes over time if later observations are more representative of the par-
ticipant’s condition. 2) Potentially Reduces Bias: Addresses some issues associated
with LOCF by using future data, which may be more accurate. On the other hand,
the disadvantages of NOCB method are: 1) Assumption of Continuity: Assumes
that the value observed in the future can be reliably applied to the past, which may
not always be valid. 2) Temporal Distortion: Can introduce bias if there are sys-
tematic changes between the missing data point and the next observation. 3) More
Complex: Generally considered less intuitive and harder to justify in some con-
texts than LOCF.

4.1.4. Hot Deck Imputation

Hot deck imputation involves replacing missing values with observed values from
similar respondents or cases [35]. This method can help preserve the distribution
of the data but may be challenging to implement for large datasets. This method
replaces missing values with values from a similar donor record (a record with
non-missing values) in the dataset.
The main steps of Hot Deck Imputation are: 1) Define a set of matching criteria

DOI: 10.4236/jcc.2024.1211004 60 Journal of Computer and Communications

 M. Alwateer et al.

(e.g., age, gender, income) based on the variables with available data. 2) For each
missing value, find a donor record that matches the criteria. 3) Replace the miss-
ing value with the corresponding value from the donor record. Advantages: Sim-
ple, can be effective for handling missing values in categorical variables.

4.1.5. Multivariate Imputation by Chained Equations (MICE)

This method generates n imputed datasets [36]. Parameter estimates and stand-
ard errors are calculated for each dataset, and then pooled to obtain overall esti-
mates ( P ) and variances ( V ):
1 n
P= ∑ Pi
n i =1

1 n
V = ∑Vi
n i =1

Between-dataset variability ( B ) is also calculated:

1 n
( Pi − P )
2
=B ∑
n − 1 i =1

4.1.6. Neighborhood Aware Autoencoder (NAA)

This approach uses a denoising autoencoder, pre-imputed with kNN ( k = 5 ), to
learn feature relationships and impute missing values [37]. The encoder and de-
coder are defined by:

(
f enc ( X ) = s X ⋅ W T + b )
(
f dec ( Y ) = s Y ⋅ W T + b )
where Y = f enc ( X ) and Z = f dec ( Y ) are the hidden and output vectors, re-
spectively; W and b are the encoder weights and bias; and W and b are
the decoder weights and bias. Training minimizes the reconstruction error be-
tween X and Z .

4.1.7. Improved Neighborhood Aware Autoencoder (I-NAA)

This enhanced version uses an undercomplete autoencoder architecture. To avoid
overfitting to the initial kNN imputation, the kNN imputation is updated every
10 epochs, varying the k value within a predefined range. Furthermore, the
missing values to be imputed are randomly selected at the start of each epoch. A
custom loss function combines mean squared error (MSE) for numerical features
and binary cross-entropy (BCE) for categorical features:
1 N ˆ
( )
2
MSE
= ∑ Yi − Yi
N i =1

1 N +C
BCE =− ∑ yi ⋅ log ( p ( yi ) ) + (1 − yi ) ⋅ log (1 − p ( yi ) )
N =i N +1
Loss RMSE + BCE
=

DOI: 10.4236/jcc.2024.1211004 61 Journal of Computer and Communications

M. Alwateer et al.

4.1.8. Multiple Imputation (MI)

Multiple imputation (MI) is a powerful technique for handling missing data that
addresses the limitations of single imputation methods. Unlike single imputation,
which replaces missing values with a single estimate, MI generates multiple com-
plete datasets by imputing missing values multiple times, each time using different
plausible values [38]. This approach accounts for the uncertainty introduced by
missing data, leading to more accurate and robust analyses.
The main steps of MI method are: 1) Imputation: Multiple complete datasets
are created by imputing the missing values using a statistical model that accounts
for the relationships between variables. The model is typically based on the ob-
served data and assumes a specific distribution for the missing values. Each im-
puted dataset is generated using different random draws from the conditional dis-
tribution of the missing values, reflecting the uncertainty associated with the miss-
ing data. 2) Analysis: Each of the imputed datasets is analyzed separately using the
chosen statistical methods. This results in multiple sets of estimates for the pa-
rameters of interest. 3) Pooling: The results from each imputed dataset are com-
bined using appropriate methods to obtain a single set of estimates and standard
errors that reflect the uncertainty introduced by missing data. The most common
pooling methods include averaging the estimates and variances across the im-
puted datasets.
The advantages of Multiple Imputation are: 1) Accounts for Uncertainty: MI
explicitly acknowledges the uncertainty associated with missing values by gener-
ating multiple plausible estimates. This results in more realistic confidence inter-
vals and p-values [39]. 2) Reduces Bias: By generating multiple imputed datasets,
MI reduces the bias introduced by single imputation methods, especially when the
missing data is not missing at random. 3) More Accurate Estimates: MI generally
produces more accurate estimates of parameters and statistical tests than single
imputation methods. 4) Provides Insights into Missing Data: The variability of
estimates across imputed datasets can provide insights into the sensitivity of the
analysis to the missing data.
The challenges of Multiple Imputation are: 1) Computational Complexity: MI
can be computationally intensive, especially for large datasets and complex mod-
els. 2) Model Selection: Choosing the appropriate imputation model is crucial.
The model should accurately reflect the relationships between variables and the
distribution of the missing data. 3) Software Requirements: Specialized software
is often required to perform multiple imputation, as it involves generating and
analyzing multiple datasets.
The choice of imputation method depends on the specific characteristics of the
data, the nature of the missing data, and the goals of the analysis. It is important
to consider the potential biases and limitations of each method before applying it
to your data. Table 2 summarizing common imputation methods, their use cases,
advantages, disadvantages, Python packages, and suitability for classification or
regression problems.

DOI: 10.4236/jcc.2024.1211004 62 Journal of Computer and Communications

 M. Alwateer et al.

Table 2. Summary of imputation methods.

Method Use cases Advantages Disadvantages Python package Problem type

Simple missing value Can introduce bias,

replacement, suitable Simple, especially for non-normally
Mean/Median “SimpleImputer”
for numerical data computationally distributed data. Does not Regression
Imputation (scikit-learn)
with a clear central inexpensive. account for relationships
tendency. between variables.

Handles both numerical Accounts for

Can be computationally
K-Nearest and categorical data, relationships between
expensive for large “KNNImputer” Regression/
Neighbors accounts for variables, effective for
datasets, sensitive to (scikit-learn) Classification
(KNN) relationships between both numerical and
the choice of k.
variables. categorical data.

Primarily for time Can introduce bias if the

Last Simple, can be
series data, replaces data is not trending, can
Observation effective for time “fillna (method =
missing values with propagate errors if there Time Series
Carried Forward series data with a ‘ffill’)” (pandas)
the last observed are consecutive missing
(LOCF) strong trend.
value. values.

Accounts for
Handles complex
uncertainty in
Multiple missing data patterns, Can be computationally “IterativeImputer”
imputation, can provide Regression/
Imputation accounts for expensive, requires (scikit-learn),
more accurate estimates Classification
(MI) uncertainty in specialized software. “fancyimpute”
than single imputation
imputation.
methods.

Primarily for
Simple, can be Can introduce bias if the
categorical data, “KNNImputer”
Hot-Deck effective for handling donor records are not truly
replaces missing values (scikit-learn) can Classification
Imputation missing values in similar to the record with
with values from a be adapted
categorical variables. the missing value.
similar donor record.

4.2. Advanced Imputation Techniques

Regression imputation uses the relationship between variables to predict missing
values based on observed data [9]. This method can account for relationships be-
tween variables but may overestimate the strength of these relationships.

4.2.1. K-Nearest Neighbors (KNN) Imputation

This method identifies the k-nearest neighbors to the missing value based on the
similarity of other features [40]. The missing value is then replaced with the aver-
age (for numerical data) or the most frequent value (for categorical data) among
those neighbors.
The main steps of KNN Imputation are: 1) Calculate the distance between the
data point with the missing value and all other points in the dataset. 2) Identify
the k-nearest neighbors based on these distances. 3) For numerical data, compute
the average of the corresponding values in the k-nearest neighbors and use it as
the imputed value. For categorical data, select the most frequent value among the

DOI: 10.4236/jcc.2024.1211004 63 Journal of Computer and Communications

M. Alwateer et al.

neighbors.
Missing values were imputed using the average (numerical features) or mode
(categorical features) of the k nearest neighbors in feature space, using Euclid-
ean distance. For example, k = 4 . Formally, for a sample S ( X , Y ,0 ) with four
nearest neighbors
= N 4 {=
( X i , Yi ,1) | i 1, 2,3, 4} , the imputed value Y is calcu-
lated as:
arg max
 { }
z ∑ ( X i ,Yi ,1)∈N 4 1( Yi = z ) if Y is categorical
Y =
1
 ∑ i4=1Yi if Y is numerical
4
where z ∈ {0,1} and 1(Yi = z ) is an indicator function. The Euclidean distance
between points x and y is defined as:

d ( x, y ) ∑ i=1 ( xi − yi )
n 2
=

where n is the number of features.

KNN imputation identifies the k most similar cases to the ones with missing
data and uses their values for imputation [41] [42]. This method can capture com-
plex relationships in the data but may be computationally expensive for large da-
tasets. K-Nearest Neighbors (KNN) imputation is a popular method for handling
missing data, leveraging the similarities between observations. Here is a closer
look at how KNN imputation works, its advantages, and its limitations.
The advantages of KNN Imputation are: 1) Flexibility: KNN can be applied to
both numerical and categorical data, making it versatile. 2) Local Information: By
considering the closest observations, KNN can capture local data patterns, poten-
tially leading to more accurate imputations. 3) Non-parametric: KNN does not
assume a specific data distribution, which can be advantageous in real-world da-
tasets.
The limitations of KNN Imputation are: 1) Computationally Intensive: KNN
can be slow, especially with large datasets, since it requires distance calculations
for each observation. 2) Curse of Dimensionality: As the number of features in-
creases, the concept of “closeness” can become less meaningful, making it harder
to identify true neighbors. Sensitive to Outliers: The presence of outliers can skew
distance calculations, leading to poor imputation results.

4.2.2. Decision Trees and Random Forests

These methods use tree-based models to predict missing values based on other
variables [16]. They can handle both categorical and continuous variables and
capture non-linear relationships. Decision Trees and Random Forests are power-
ful machine learning techniques that can also be used to input missing values into
datasets. Here’s a breakdown of how these methods work for imputation, along
with their advantages and disadvantages.
The advantages of Decision Trees Imputation: 1) Captures Non-linear Relation-
ships: Decision trees can model complex relationships between features, potentially
leading to more accurate imputations. 2) Interpretable: The model is relatively

DOI: 10.4236/jcc.2024.1211004 64 Journal of Computer and Communications

 M. Alwateer et al.

easy to interpret, as you can visualize how decisions are made. The disadvantages
of Decision Trees Imputation: 1) Overfitting: Decision trees can easily overfit to
the training data, especially if not properly pruned. 2) Sensitivity to Noise: Outli-
ers can affect the structure of the tree, impacting the imputation results.
The advantages of Random Forests Imputation: 1) Improved Accuracy: Ran-
dom forests generally provide better accuracy than single decision trees due to
their ensemble nature, reducing overfitting and variance. 2) Robust to Outliers:
The averaging mechanism makes random forests less sensitive to outliers com-
pared to individual trees. 3) Handles Large Datasets: Random forests can effec-
tively manage large datasets with high dimensionality. The disadvantages of Ran-
dom Forests Imputation: 1) Complexity: The model is less interpretable than a
single decision tree, as it’s harder to visualize how predictions are made. 2) Com-
putationally Intensive: Training multiple trees can be resource-intensive, espe-
cially for large datasets.

4.2.3. Support Vector Machines (SVM)

Support Vector Machines (SVM) are primarily known for classification and re-
gression tasks. However, they can also be utilized to input missing data. SVM-
based imputation methods use support vector regression to predict missing values
[43] [44]. These methods can be effective for high-dimensional data but may re-
quire careful tuning of hyperparameters.
The advantages of SVM-Based Imputation are: 1) Effective for Non-linear Re-
lationships: SVM can capture complex, non-linear relationships in the data by us-
ing different kernel functions (e.g., polynomial, radial basis function). 2) Robust-
ness to Overfitting: SVM includes regularization parameters that help prevent
overfitting, making it suitable for high-dimensional datasets. 3) Flexibility: SVM
can be applied to both classification (categorical variables) and regression (con-
tinuous variables) tasks.
The limitations of SVM-Based Imputation are: 1) Computational Complexity:
SVM can be computationally intensive, particularly for large datasets, due to the
optimization required for finding the best hyperplane. 2) Parameter Sensitivity:
The performance of SVM can be sensitive to the choice of kernel and hyperpa-
rameters (e.g., C and gamma), requiring careful tuning. 3) Requires Sufficient
Data: SVM models generally require a substantial amount of complete data to
build an accurate model, which may not always be available.

4.3. Deep Learning Approaches

Autoencoders are neural networks that can learn compressed representations of
data and have been applied to missing data imputation [45]-[47]. They can cap-
ture complex patterns in the data but may require large amounts of training data.

Generative Adversarial Networks (GANs)

GANs have been adapted for missing data imputation by learning to generate re-
alistic imputed values [48] [49]. This approach can produce high-quality imputa-

DOI: 10.4236/jcc.2024.1211004 65 Journal of Computer and Communications

M. Alwateer et al.

tions but may be challenging to train and tune. The Generative Adversarial Im-
putation Network (GAIN) [50] employs a generative adversarial network (GAN)
architecture. Unlike standard GANs, the discriminator in GAIN does not classify
the entire generated output as real or fake; instead, it classifies each individual
variable as either imputed or observed. Convergence is achieved when the gener-
ator produces imputations indistinguishable from the true data distribution. A
“hint” mechanism augments the discriminator’s input with partial information
about the missing values (M), represented by the hint vector H. This hint is typi-
cally a proportion of M (e.g., 90% identical). The generator then learns to impute
the remaining values. The original work demonstrates that insufficient hints lead
to multiple optimal generator outputs.
Formally, given a random vector Z, the generator produces an imputed dataset
X i , from which X i is derived (Equation (1)). The discriminator loss function,
LD , is defined as:
,H )
LD ( M , M= ∑  M i ⋅ log ( M i ) + (1 − M i ) ⋅ log (1 − M i ) 
i:H i =0

where M is the true mask vector, M is the generated mask vector, and H is
the hint vector. The summation is restricted to indices where H i = 0 to prevent
overfitting to the hint. The discriminator is trained to minimize this loss:
batchsize
min D − ∑ LD ( M i , M i , H i )
i =1

The generator loss function comprises two terms: LG , which measures the
generator’s ability to deceive the discriminator, and LM , which quantifies the ac-
curacy of the imputation for observed values:

LG ( M , M , H ) =
− ∑ (1 − M i ) ⋅ log ( M i )
i:H i =0

d
(
LM X , X = )
−∑ M i ⋅ Diff X i , X i ( )
i =1

where d is the data dimensionality and Diff is defined as:

) ( )
 X − X 2 if X i is numerical

(

Diff X , X =  i i

− X i log ( X i ) if X i is binary

The generator is trained to minimize the combined loss:

batchsize
min G ∑ LG ( M i , M i , H i ) + α LM i X i , X i( )
i =1

where α is a scaling parameter.

4.4. Time Series-Specific Methods

ARIMA Models
Autoregressive Integrated Moving Average (ARIMA) models are a class of statis-
tical models used for analyzing and forecasting time series data. ARIMA models

DOI: 10.4236/jcc.2024.1211004 66 Journal of Computer and Communications

 M. Alwateer et al.

can be employed to impute missing values in time series data by leveraging tem-
poral dependencies [51] [52]. They are particularly useful when the data exhibits
trends and seasonality. An ARIMA model is denoted as ARIMA (p, d, q), where
“p” represents the order of the autoregressive (AR) component, “d” represents the
degree of difference required to make the time series stationary, and “q” repre-
sents the order of the moving average (MA) component. The AR component
models the relationship between the current observation and previous observa-
tions; the MA component models the relationship between the current observa-
tion and past forecast errors, and differencing (d) removes trends and makes the
series stationary. A general ARIMA (p, d, q) model can be represented by the fol-
lowing equation:
φ ( B )(1 − B ) yt =
θ ( B ) t
d

where yt is the time series at time t , B is the backshift operator ( Byt = yt −1 ),

φ ( B ) is the autoregressive polynomial of order p , θ ( B ) is the moving aver-
age polynomial of order q , and t is white noise. The choice of p, d , and q
values is crucial for model fitting and depends on the characteristics of the specific
time series being analyzed. Techniques like autocorrelation and partial autocor-
relation functions (ACF and PACF) are often used to identify suitable model or-
ders. Table 3 shows the summary of advanced imputation techniques.

Table 3. Summary of advanced imputation techniques.

Method Advantages Disadvantages Python package Problem type

Multiple Imputation Handles complex relationships Can be computationally “IterativeImputer”

Regression/
by Chained Equations between variables, accounts for expensive, requires careful (scikit-learn),
Classification
(MICE) uncertainty in imputation. model selection. “fancyimpute”

Handles mixed-type data Can be computationally

Random Forest Regression/
(numerical and categorical), expensive, may overfit if the “MissForest”
Imputation Classification
robust to outliers. data is highly correlated.

Can generate realistic synthetic Requires significant

Generative Adversarial “Tensorflow”, Regression/
data, handles complex data computational resources, can
Networks (GANs) “Pytorch” Classification
distributions. be challenging to train.

Can capture complex non-linear Requires large amounts of

Deep Learning “Tensorflow”, Regression/
relationships in the data, handles data, can be computationally
Imputation “Pytorch” Classification
high-dimensional datasets. expensive, may overfit.

Accounts for prior knowledge Can be computationally

Bayesian “PyMC3”, Regression/
and uncertainty, provides intensive, requires careful
Imputation “PyStan” Classification
probabilistic estimates. model specification.

5. Evaluation Metrics for Imputation Methods

Assessing the performance of imputation methods is crucial for selecting appro-
priate techniques [53]. Evaluating the performance of imputation methods is

DOI: 10.4236/jcc.2024.1211004 67 Journal of Computer and Communications

M. Alwateer et al.

crucial to ensure that the imputed data maintains the integrity and reliability of
the original data. Several metrics are commonly used to assess the effectiveness of
imputation techniques. The choice of evaluation metric depends on the specific
objective of the imputation and the nature of the data. These evaluation metrics
can be broadly categorized into two groups, as shown in Table 4.

Table 4. Regression and classification metrics.

Regression metrics Classification metrics

1 n TP + TN
∑ ( yi − yˆi )
2
Mean Squared Error (MSE) MSE
= Accuracy Accuracy =
n i =1 TP + TN + FP + FN

Root Mean Squared Error 1 n TP

∑ ( yi − yˆi )
2
RMSE
= Precision Precision =
(RMSE) n i =1
TP + FP

1 n TP
Mean Absolute Error (MAE) MAE
= ∑ yi − yˆi
n i =1
Recall (Sensitivity) Recall =
TP + FN

Mean Absolute Percentage 1 n yi − yˆ i Precision ∗ Recall

Error (MAPE)
=MAPE ∑
n i =1 yi
× 100% F1-Score F1-Score= 2 ∗
Precision + Recall

6. Challenges and Considerations

While imputation methods offer valuable tools for handling missing data, several
challenges and considerations must be addressed when implementing missing
data imputation to ensure effective and reliable results [54].
Bias Introduction: Imputation methods can introduce bias into the data, par-
ticularly if the missing values are not missing at random (MAR). This means that
the missingness is related to the value of the missing variable itself or other varia-
bles in the dataset. Example: If missing values in income are more likely to occur
for individuals with lower incomes, simply replacing them with the mean income
will underestimate the true average income.
Data Distribution: Imputation methods often assume that the data follows a
specific distribution (e.g., normal distribution). If the data deviates significantly
from this assumption, the imputed values may not be representative. Example:
Using mean imputation on a skewed distribution will result in imputed values that
are biased towards the tail of the distribution.
Missing Value Patterns: The pattern of missing values can significantly impact
the effectiveness of imputation methods. If missing values are clustered or follow
a specific pattern, simple methods like mean imputation may not be appropriate.
Example: If consecutive values are missing in a time series, LOCF or NOCB may
introduce significant bias.
Computational Complexity: Some imputation methods, like multiple imputa-
tion or KNN imputation, can be computationally expensive, especially for large
datasets. 1) Domain Knowledge: Incorporating domain knowledge into the impu-
tation process can significantly improve the accuracy and relevance of the imputed

DOI: 10.4236/jcc.2024.1211004 68 Journal of Computer and Communications

 M. Alwateer et al.

values. Example: In medical data, understanding the relationships between different

variables and the potential causes of missing values can guide the choice of impu-
tation method.
2) Model Selection: Choosing the appropriate imputation model is crucial. The
choice should be based on the characteristics of the data, the nature of the missing
values, and the goals of the analysis. 3) Interpretability: The interpretability of the
imputed values is important for understanding the results of the analysis.

7. Case Studies and Applications

This section illustrates the practical implementation of missing data imputation
techniques in a range of different fields. Examples of these applications can be
found as follows.
 Healthcare: Examining approaches for resolving incomplete entries in elec-
tronic health records, with a focus on maintaining data integrity and improv-
ing diagnostic accuracy [55]-[57].
 Finance: Analysis of strategies for managing incomplete datasets in financial
forecasting, specifically focusing on stock market predictions and their impli-
cations for investment decision-making [58] [59].
 Social Sciences: Investigation of techniques to mitigate the impact of non-re-
sponse in survey data, exploring methods to preserve statistical validity and
minimize bias in population-level inferences [60] [61].

8. Future Directions and Open Problems

Several areas for future research and development in missing data imputation in-
clude [62] [63].
1) Emerging techniques and research areas: Federated learning for privacy-
preserving imputation [64] [65]. Reinforcement learning for adaptive imputation
strategies [66]. Transfer learning for imputation in low-resource settings [67] [68].
2) Integration with big data and real-time systems: Developing scalable and
efficient imputation methods for streaming data and large-scale datasets [69]-[71]
as follows: Distributed Algorithms: Use scalable imputation algorithms that can
handle large datasets efficiently. Techniques like mini-batch processing or parallel
computing can be useful. Big Data Frameworks: Leverage tools like Apache Spark
or Hadoop, which can process large volumes of data quickly and support machine
learning libraries for imputation.
3) Ethical considerations in data imputation: Addressing potential biases and
fairness issues in imputation methods, especially in sensitive applications like
healthcare and criminal justice [72]-[74]. Here are some key points for these Eth-
ics: a) Transparency: Researchers should clearly communicate how missing data
will be handled, including the imputation methods used. This transparency builds
trust and allows for reproducibility. b) Bias and Misrepresentation: Imputation
can introduce bias if not done carefully. Researchers must consider whether the
imputed data accurately reflects the underlying population or if it skews results.

DOI: 10.4236/jcc.2024.1211004 69 Journal of Computer and Communications

M. Alwateer et al.

c) Informed Consent: Participants should be informed about how their data, in-
cluding any imputed values, will be used in research. This includes potential im-
plications for privacy and the integrity of their responses. d) Appropriateness of
Methods: Different imputation methods (mean, median, predictive modeling,
etc.) have different assumptions. Choosing the right method is crucial to avoid
distorting the data and the conclusions drawn from it. e) Impact on Decision-
Making: The results derived from imputed data can influence policy or clinical
decisions. Researchers should ensure that their imputation practices do not lead
to harmful outcomes. f) Equity: Consider whether the imputation methods used
could disproportionately affect certain groups. Ensuring that imputation methods
do not reinforce existing inequalities is vital. g) Ethical Oversight: It’s beneficial
to have an ethical review process in place to assess the imputation strategies and
their potential implications for participants and broader societal contexts. h) Data
Integrity: Strive to maintain the integrity of the original dataset. Imputation
should not compromise the authenticity of the data, and researchers should be
mindful of the limitations that come with imputed values. i) Training and Exper-
tise: Ensure that those involved in the imputation process have the necessary
training and understanding of the ethical implications of their work.

9. Conclusion
As data collection and analysis continue to grow in importance across various do-
mains, the field of missing data imputation is likely to see further advancements
and innovations. This review has provided a comprehensive overview of missing
data imputation techniques, from traditional methods, and statistical methods to
advanced machine learning approaches. Key observations highlight: 1) The criti-
cal role of understanding missing data mechanisms. 2) The trade-offs between
simple and complex imputation methods. 3) There is a need for careful evaluation
and selection of imputation techniques. and 4) The potential of machine learning
and deep learning approaches for handling complex missing data patterns. Future
research should focus on developing more robust, efficient, and adaptable impu-
tation methods that can handle the increasing complexity and scale of modern
datasets while addressing ethical concerns and preserving data integrity.

Conflicts of Interest
The authors declare no conflicts of interest regarding the publication of this paper.

References
[1] Enders, C.K. (2022) Applied Missing Data Analysis. Guilford Publications.
[2] Mitchel, J.T., Kim, Y.J., Choi, J., Park, G., Cappi, S., Horn, D., et al. (2011) Evaluation
of Data Entry Errors and Data Changes to an Electronic Data Capture Clinical Trial
Database. Drug Information Journal, 45, 421-430.
[Link]
[3] Schafer, J.L. and Graham, J.W. (2002) Missing Data: Our View of the State of the Art.
Psychological Methods, 7, 147-177. [Link]

DOI: 10.4236/jcc.2024.1211004 70 Journal of Computer and Communications

 M. Alwateer et al.

[4] Little, R. and Rubin, D. (2019) Statistical Analysis with Missing Data. Third Edition,
Wiley. [Link]
[5] Lazar, N.A. (2003) Statistical Analysis with Missing Data. Technometrics, 45, 364-
365. [Link]
[6] Hajjar, S. (2018) Statistical Analysis: Internal-Consistency Reliability and Construct
Validity. International Journal of Quantitative and Qualitative Research Methods, 6,
27-38.
[7] Noor, T.H., Atlam, E., Almars, A.M., Noor, A. and Malki, A.S. (2023) An IoT-Based
Energy Conservation Smart Classroom System. Intelligent Automation & Soft Com-
puting, 35, 3785-3799. [Link]
[8] Zhan, Y., Xia, Y., Liu, Y., Li, F. and Wang, Y. (2017) Incentive-Aware Time-Sensitive
Data Collection in Mobile Opportunistic Crowdsensing. IEEE Transactions on Ve-
hicular Technology, 66, 7849-7861. [Link]
[9] Molenberghs, G. and Verbeke, G. (2013) Missing Data. In: Scott, M.A., Simonoff, J.S.
and Marx, B.D., Eds., The SAGE Handbook of Multilevel Modeling, SAGE Publica-
tions Ltd, 403-424. [Link]
[10] Graham, J.W. (2009) Missing Data Analysis: Making It Work in the Real World. An-
nual Review of Psychology, 60, 549-576.
[Link]
[11] Sterne, J.A.C., White, I.R., Carlin, J.B., Spratt, M., Royston, P., Kenward, M.G., et al.
(2009) Multiple Imputation for Missing Data in Epidemiological and Clinical Re-
search: Potential and Pitfalls. BMJ, 338, b2393-b2393.
[Link]
[12] Masconi, K.L., Matsha, T.E., Echouffo-Tcheugui, J.B., Erasmus, R.T. and Kengne,
A.P. (2015) Reporting and Handling of Missing Data in Predictive Research for Prev-
alent Undiagnosed Type 2 Diabetes Mellitus: A Systematic Review. EPMA Journal, 6,
Article No. 7. [Link]
[13] Ferris, J.A. (2009) Missing Data: A Gentle Introduction. Drug and Alcohol Review,
28, 90-91. [Link]
[14] Enders, C.K. (2017) Multiple Imputation as a Flexible Tool for Missing Data Han-
dling in Clinical Research. Behaviour Research and Therapy, 98, 4-18.
[Link]
[15] Bertsimas, D., Pawlowski, C. and Zhuo, Y.D. (2018) From Predictive Methods to
Missing Data Imputation: An Optimization Approach. Journal of Machine Learning
Research, 18, 1-39.
[16] Stekhoven, D.J. and Bühlmann, P. (2011) Missforest—Non-Parametric Missing
Value Imputation for Mixed-Type Data. Bioinformatics, 28, 112-118.
[Link]
[17] Van Buuren, S. (2018) Flexible Imputation of Missing Data. CRC Press.
[18] Rubin, D.B. (1976) Inference and Missing Data. Biometrika, 63, 581-592.
[Link]
[19] Doreswamy, Gad, I. and Manjunatha, B.R. (2017) Performance Evaluation of Predic-
tive Models for Missing Data Imputation in Weather Data. 2017 International Con-
ference on Advances in Computing, Communications and Informatics (ICACCI),
Udupi, 13-16 September 2017, 1327-1334.
[Link]
[20] Hosahalli, D. and Gad, I. (2018) A Generic Approach of Filling Missing Values in
NCDC Weather Stations Data. 2018 International Conference on Advances in

DOI: 10.4236/jcc.2024.1211004 71 Journal of Computer and Communications

M. Alwateer et al.

Computing, Communications and Informatics (ICACCI), Bangalore, 19-22 Septem-

ber 2018, 143-149. [Link]
[21] Khanani, N. (2021) Addressing Missing Data in Educational Evaluation: Predictive
Mean Matching Imputation for Test Score Data. Proceedings of the 2021 AERA An-
nual Meeting, 4 September 2021, 15. [Link]
[22] Thakur, S., Choudhary, J. and Singh, D.P. (2021) A Survey on Missing Values Han-
dling Methods for Time Series Data. In: Sheth, A., Sinhal, A., Shrivastava, A. and
Pandey, A.K., Eds., Intelligent Systems, Springer Singapore, 435-443.
[Link]
[23] Psychogyios, K., Ilias, L., Ntanos, C. and Askounis, D. (2023) Missing Value Imputa-
tion Methods for Electronic Health Records. IEEE Access, 11, 21562-21574.
[Link]
[24] Jiménez, O., Jesús, A. and Wong, L. (2023) Model for the Prediction of Dropout in
Higher Education in Peru Applying Machine Learning Algorithms: Random Forest,
Decision Tree, Neural Network and Support Vector Machine. 2023 33rd Conference
of Open Innovations Association (FRUCT), Zilina, 24-26 May 2023, 116-124.
[Link]
[25] Nida, H. (2023) Comparison of Missing Data Imputation Methods Using Weather
Data. Pakistan Journal of Agricultural Sciences, 60, 327-336.
[Link]
[26] Teegavarapu, R.S.V. (2024) Applications: Imputation of Missing Hydrometeorologi-
cal Data. Water Science and Technology Library, 108, 491-517.
[Link]
[27] Almeida, A., Brás, S., Sargento, S. and Pinto, F.C. (2024) Focalize K-NN: An Imputa-
tion Algorithm for Time Series Datasets. Pattern Analysis and Applications, 27, Ar-
ticle No. 39. [Link]
[28] Kowsar, I., Rabbani, S.B. and Samad, M.D. (2024) Attention-Based Imputation of
Missing Values in Electronic Health Records Tabular Data. 2024 IEEE 12th Interna-
tional Conference on Healthcare Informatics (ICHI), Orlando, 3-6 June 2024, 177-
182. [Link]
[29] Little, R.J.A. (1988) A Test of Missing Completely at Random for Multivariate Data
with Missing Values. Journal of the American Statistical Association, 83, 1198-1202.
[Link]
[30] Hashim, H., Almaliki, M., El-Agamy, R., El Sharkasy, M., Dagnew, G., Gad, I.,
Ghoneim, O., et al. (2021) Integrating Data Warehouse and Machine Learning to
Predict on COVID-19 Pandemic Empirical Data. Journal of Theoretical and Applied
Information Technology, 99, 159-170.
[31] Malki, Z., Atlam, E., Hassanien, A.E., Dagnew, G., Elhosseini, M.A. and Gad, I. (2020)
Association between Weather Data and COVID-19 Pandemic Predicting Mortality
Rate: Machine Learning Approaches. Chaos, Solitons & Fractals, 138, Article ID:
110137. [Link]
[32] Laaksonen, S. (2018) Survey Methodology and Missing Data: Tools and Techniques
for Practitioners. Springer. [Link]
[33] Schafer, J.L. (1999) Multiple imputation: a primer. Statistical Methods in Medical Re-
search, 8, 3-15. [Link]
[34] Lee, Y. (2023) Imputation Method Using Local Linear Regression Based on Bidirectional
k-nearest-components. Journal of information and communication convergence engi-
neering, 21, 62-67. [Link]

DOI: 10.4236/jcc.2024.1211004 72 Journal of Computer and Communications

 M. Alwateer et al.

[35] Andridge, R.R. and Little, R.J.A. (2010) A Review of Hot Deck Imputation for Survey
Non-Response. International Statistical Review, 78, 40-64.
[Link]
[36] van Buuren, S. (2007) Multiple Imputation of Discrete and Continuous Data by Fully
Conditional Specification. Statistical Methods in Medical Research, 16, 219-242.
[Link]
[37] Aidos, H. and Tomas, P. (2021) Neighborhood-Aware Autoencoder for Missing
Value Imputation. 2020 28th European Signal Processing Conference (EUSIPCO),
Amsterdam, 18-21 January 2021, 1542-1546.
[Link]
[38] Rubin, D.B. (1987) Multiple Imputation for Nonresponse in Surveys. Wiley.
[Link]
[39] Rombach, I., Gray, A.M., Jenkinson, C., Murray, D.W. and Rivero-Arias, O. (2018)
Multiple Imputation for Patient Reported Outcome Measures in Randomised Con-
trolled Trials: Advantages and Disadvantages of Imputing at the Item, Subscale or
Composite Score Level. BMC Medical Research Methodology, 18, Article No. 87.
[Link]
[40] Murti, D.M.P., Pujianto, U., Wibawa, A.P. and Akbar, M.I. (2019) K-Nearest Neigh-
bor (K-NN) Based Missing Data Imputation. 2019 5th International Conference on
Science in Information Technology (ICSITech), Yogyakarta, 23-24 October 2019, 83-
88. [Link]
[41] Troyanskaya, O., Cantor, M., Sherlock, G., Brown, P., Hastie, T., Tibshirani, R., et al.
(2001) Missing Value Estimation Methods for DNA Microarrays. Bioinformatics, 17,
520-525. [Link]
[42] Gad, I., Elmezain, M., Alwateer, M.M., Almaliki, M., Elmarhomy, G. and Atlam, E.
(2023) Breast Cancer Diagnosis Using a Machine Learning Model and Swarm Intel-
ligence Approach. 2023 1st International Conference on Advanced Innovations in
Smart Cities (ICAISC), Jeddah, 23-25 January 2023, 1-5.
[Link]
[43] Mallinson, H. and Gammerman, A. (2003) Imputation Using Support Vector Ma-
chines. University of London Egham, UK: Department of Computer Science Royal
Holloway.
[44] Noor, T.H., Almars, A., Gad, I., Atlam, E. and Elmezain, M. (2022) Spatial Impres-
sions Monitoring during COVID-19 Pandemic Using Machine Learning Techniques.
Computers, 11, Article 52. [Link]
[45] Vincent, P., Larochelle, H., Bengio, Y. and Manzagol, P. (2008) Extracting and Com-
posing Robust Features with Denoising Autoencoders. Proceedings of the 25th inter-
national conference on Machine learning—ICML’08, Helsinki, 5-9 July 2008, 1096-
1103. [Link]
[46] Noor, T.H., Almars, A.M., Atlam, E. and Noor, A. (2022) Deep Learning Model for
Predicting Consumers’ Interests of IoT Recommendation System. International Jour-
nal of Advanced Computer Science and Applications, 13, 161-170.
[Link]
[47] Elmezain, M., Malki, A., Gad, I. and Atlam, E. (2022) Hybrid Deep Learning Model-
Based Prediction of Images Related to Cyberbullying. International Journal of Ap-
plied Mathematics and Computer Science, 32, 323-334.
[Link]
[48] Shahbazian, R. and Trubitsyna, I. (2022) DEGAIN: Generative-Adversarial-Net-
work-Based Missing Data Imputation. Information, 13, Article 575.

DOI: 10.4236/jcc.2024.1211004 73 Journal of Computer and Communications

M. Alwateer et al.

[Link]
[49] Malki, A., Atlam, E. and Gad, I. (2022) Machine Learning Approach of Detecting
Anomalies and Forecasting Time-Series of IoT Devices. Alexandria Engineering
Journal, 61, 8973-8986. [Link]
[50] Yoon, J., Jordon, J. and van der Schaar, M. (2018) GAIN: Missing Data Imputation
Using Generative Adversarial Nets, Proceedings of the 35th International Conference
on Machine Learning, Stockholm, 10-15 July 2018, 5689-5698.
[Link]
[51] Chhabra, G. (2023) Comparison of Imputation Methods for Univariate Time Series.
International Journal on Recent and Innovation Trends in Computing and Commu-
nication, 11, 286-292. [Link]
[52] Malki, A., Atlam, E., Hassanien, A.E., Ewis, A., Dagnew, G. and Gad, I. (2022)
SARIMA Model-Based Forecasting Required Number of COVID-19 Vaccines Glob-
ally and Empirical Analysis of Peoples’ View towards the Vaccines. Alexandria Engi-
neering Journal, 61, 12091-12110. [Link]
[53] Collins, L.M., Schafer, J.L. and Kam, C. (2001) A Comparison of Inclusive and Re-
strictive Strategies in Modern Missing Data Procedures. Psychological Methods, 6,
330-351. [Link]
[54] Emmanuel, T., Maupong, T., Mpoeleng, D., Semong, T., Mphago, B. and Tabona, O.
(2021) A Survey on Missing Data in Machine Learning. Journal of Big Data, 8, Article
No. 140. [Link]
[55] Gustems-Carnicer, J. and Calderón, C. (2012) Coping Strategies and Psychological
Well-Being among Teacher Education Students. European Journal of Psychology of
Education, 28, 1127-1140. [Link]
[56] Atlam, E., Masud, M., Rokaya, M., Meshref, H., Gad, I. and Almars, A.M. (2024)
EASDM: Explainable Autism Spectrum Disorder Model Based on Deep Learning.
Journal of Disability Research, 3, 1-15. [Link]
[57] Masud, M., Almars, A.M., Rokaya, M.B., Meshref, H., Gad, I. and Atlam, E. (2024) A
Novel Light-Weight Convolutional Neural Network Model to Predict Alzheimer’s
Disease Applying Weighted Loss Function. Journal of Disability Research, 3, 1-10.
[Link]
[58] McMahon, P., Zhang, T. and Dwight, R.A. (2020) Approaches to Dealing with Miss-
ing Data in Railway Asset Management. IEEE Access, 8, 48177-48194.
[Link]
[59] Bennin, K.E., Tahir, A., MacDonell, S.G. and Börstler, J. (2021) An Empirical Study
on the Effectiveness of Data Resampling Approaches for Cross-Project Software De-
fect Prediction. IET Software, 16, 185-199. [Link]
[60] Hsu, C.C. and Sandford, B.A. (2019) Minimizing Non-Response in the Delphi Pro-
cess: How to Respond to Non-Response. Practical Assessment, Research, and Evalu-
ation, 12, 17.
[61] Carpenter, J.R., Roger, J.H. and Kenward, M.G. (2013) Analysis of Longitudinal Tri-
als with Protocol Deviation: A Framework for Relevant, Accessible Assumptions, and
Inference via Multiple Imputation. Journal of Biopharmaceutical Statistics, 23, 1352-
1371. [Link]
[62] Adnan, F.A., Jamaludin, K.R., Wan Muhamad, W.Z.A. and Miskon, S. (2022) A Re-
view of the Current Publication Trends on Missing Data Imputation over Three Dec-
ades: Direction and Future Research. Neural Computing and Applications, 34,
18325-18340. [Link]

DOI: 10.4236/jcc.2024.1211004 74 Journal of Computer and Communications

 M. Alwateer et al.

[63] Pedersen, A., Mikkelsen, E., Cronin-Fenton, D., Kristensen, N., Pham, T.M.,
Pedersen, L., et al. (2017) Missing Data and Multiple Imputation in Clinical Epide-
miological Research. Clinical Epidemiology, 9, 157-166.
[Link]
[64] Li, T., Sahu, A.K., Talwalkar, A. and Smith, V. (2020) Federated Learning: Challenges,
Methods, and Future Directions. IEEE Signal Processing Magazine, 37, 50-60.
[Link]
[65] Singh, P., Singh, M.K., Singh, R. and Singh, N. (2022) Federated Learning: Chal-
lenges, Methods, and Future Directions. In: Yadav, S.P., Bhati, B.S., Mahato, D.P. and
Kumar, S., Eds., Federated Learning for IoT Applications, Springer International
Publishing, 199-214. [Link]
[66] Ma, C., Tschiatschek, S., Palla, K., Hernández-Lobato, J.M., Nowozin, S. and Zhang,
C. (2018) Eddi: Efficient Dynamic Discovery of High-Value Information with Partial
Vae. arXiv: 1809.11142.
[67] Sultana, Z., Akter, S. and Yeasmin, A. (2022) Transfer Learning Approach Applied to
Data Imputation. University of Liberal Arts Bangladesh.
[Link]
[68] Lyu, L., Hu, Y., Wang, N., Zhou, X. and Fang, M. (2022) Application of Deep Learn-
ing and Transfer Learning in Continuous Missing Value Imputation of Water Quality
Data. 2022 IEEE 8th International Conference on Computer and Communications
(ICCC), Chengdu, 9-12 December 2022, 1211-1216.
[Link]
[69] Anagnostopoulos, I. (2018) Fintech and Regtech: Impact on Regulators and Banks.
Journal of Economics and Business, 100, 7-25.
[Link]
[70] Gupta, M. and Gupta, B. (2020) A New Scalable Approach for Missing Value Impu-
tation in High-Throughput Microarray Data on Apache Spark. International Journal
of Data Mining and Bioinformatics, 23, 79-100.
[Link]
[71] Petrozziello, A., Jordanov, I. and Sommeregger, C. (2018) Distributed Neural Net-
works for Missing Big Data Imputation. 2018 International Joint Conference on Neu-
ral Networks (IJCNN), Rio de Janeiro, 8-13 July 2018, 1-8.
[Link]
[72] Rajkomar, A., Hardt, M., Howell, M.D., Corrado, G. and Chin, M.H. (2018) Ensuring
Fairness in Machine Learning to Advance Health Equity. Annals of Internal Medi-
cine, 169, 866-872. [Link]
[73] Chandler, R.K., Fletcher, B.W. and Volkow, N.D. (2009) Treating Drug Abuse and
Addiction in the Criminal Justice System. JAMA, 301, 183-190.
[Link]
[74] De Pau, M., Vruggink, R., Vandevelde, S. and Vander Laenen, F. (2023) Culturally
Sensitive Forensic Mental Healthcare for Racialized People Labeled as Not Criminally
Responsible: A Scoping Review. International Journal of Forensic Mental Health, 22,
276-288. [Link]

DOI: 10.4236/jcc.2024.1211004 75 Journal of Computer and Communications