Decision Support Systems 184 (2024) 114276

Contents lists available at ScienceDirect

Decision Support Systems

journal homepage: [Link]/locate/dss

Explainable AI for enhanced decision-making

A R T I C L E I N F O A B S T R A C T

Keywords This paper contextualizes explainable artificial intelligence (AI) for enhanced decision-making and serves as an
Explainable artificial intelligence editorial for the corresponding special issue. AI is defined as the development of computer systems that are able
Interpretability to perform tasks that normally require human intelligence by understanding, processing, and analyzing large
Visualizations
amounts of data. AI has been a dominant domain for several decades in the information systems (IS) literature.
To this end, we define explainable AI (XAI) as the process that allows one to understand how an AI system
decides, predicts, and performs its operations. First, we contextualize its current role for improved business
decision-making. Second, we discuss three underlying dimensions of XAI that serve as broader innovation
grounds to make better and more informed decisions, i.e., data, method, and application. For each of the
contributing papers in this special issue, we describe their major contributions to the field of XAI for decision
making. In conclusion, this paper further presents a future research agenda for IS researchers in the XAI field.

1. Introduction that allows us to understand how an AI system decides, predicts, and

performs its operations. Therefore, XAI reveals the strengths and
Artificial Intelligence (AI) defined as the development of computer weaknesses of the decision-making strategy and explains the rationale of
systems that are able to perform tasks normally requiring human intel­ the decision support system (DSS) [19]. Numerous scholars confirm that
ligence by understanding, processing, and analyzing large amounts of XAI is the key to developing and implementing AI in industries such as
data has been a prevalent domain in information systems (IS) for several retail, banking and financial services, healthcare, manufacturing and
decades now. An increasing number of businesses rely on AI to achieve supply chain [18,20–22]. In addition, XAI has also received the attention
outcomes that operationally and/or strategically support (human) de­ of governments due to its ability to improve the efficiency and effec­
cision making in different domains. At present, AI-based machine tiveness of government functionalities and decision supports [23].
learning (ML) has become widely popular as a subfield of AI, both in In fact, in many cases, understanding why a model makes certain
academia as in industry. ML has been widely used to enhance decision decisions and predictions is as important as its accuracy. Because model
making, including predicting organ transplantation risk [1], forecasting explainability helps managers better understand model parameters and
remaining useful life of machinery [2], student dropout prediction [3], apply them more confidently, allowing managers to communicate the
disentangling human trafficking types [4], money laundering detection analytical rationale more convincingly for their decisions to stake­
[5,6], among others. In the early days, AI attempted to imitate human holders [24]. Among others, exploring the drivers of AI explainability
decision-making rules with only partial success, as humans often could and precise understandability in decision making is one of the main
not accurately describe the decision-making rules they use to solve contributions of this special issue. In concrete terms, we explore three
problems [7]. With the development of advanced AI, exciting progress main domains of innovation that make up an XAI strategy that helps to
has been made in algorithmic development to support decision making ensure transparency during decision-making, that is, data, method and
in various fields including finance [8,9], marketing [10–12], education application.
[3], human resource management [13], tourism [14], aviation [15] and Section 2 zooms in on the dimensions of XAI for enhanced decision
others [16]. making, while Section 3 provides an overview and summarizes the
Recently, advances have heavily focused on boosting the predictive contributions included in this special issue. In addition, Section 4 con­
accuracy of AI methods, with deep learning (DL) methods being a cludes this article and discusses interesting future research directions in
prevalent example. For instance, [17] use the text on corporate websites XAI for the IS field.
to improve business failure identification using advanced DL-based
natural language processing (NLP) methods. The stringent focus on
improved prediction performance often comes at the expense of reduced
explainability, which leads to decision makers’ distrust and even
rejection of AI systems [18]. Explainable AI (XAI) describes the process

[Link]

Available online 28 June 2024

0167-9236/© 2024 Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
K. Coussement et al. Decision Support Systems 184 (2024) 114276

2. Dimensions of explainable AI for enhanced decision-making benchmarking studies have reinforced this perception, with ensemble
methods, particularly XGBoost, garnering significant attention in the
2.1. Data recent IS literature. [32] demonstrated the dominance of XGBoost in
effectively predicting financial risks in supply chain finance. [33] find
The link between XAI and different types of data, including obser­ favorable results for XGBoost compared to other ML models in pre­
vational data with structured and unstructured data, as well as synthetic dicting the probability of risk and the frequency of claims in the context
data, is a vital aspect of the interpretability and transparency of the ML of automobile insurances. However, the trade-off is that these black-box
outcomes. Extant IS literature has already dived into often manually models often compromise interpretability.
crafted variables from structured data that are inherently interpretable However, recent advances have been made to elucidate the inner
and directly improve managerial decision-making. For example, [25] workings of black-box models. Post-hoc interpretation tools such as
designed direct understandable variables – linked to the user network, SHAP [34], ICE [35], and LIME [36] now allow the examination of the
user behavior, message readability, and message structure – out of X impact of various drivers on final predictions. For example, [37] used
social media posts to improve the identification of influential customers. the SHAP algorithm to provide intuitive insight into the context of
However, existing IS literature is less lenient to converting unstructured readmissions from the emergency department of COVID-19. On the
– textual, audio, or image – data into directly interpretable features, and methodological front, [28] incorporates an adapted version of the
often the focus lies on optimizing the predictive performance for Ordinal Priority Approach (OPA) into a fuzzy logic context. These
enhanced decision-making. Furthermore, experimental data plays a methodological choices aim to improve the model’s adaptability and
crucial role in XAI solutions by allowing controlled experimentation transparency, although further research would be needed to generalize
using real-life field testing or the creation of synthetic datasets (e.g. these methods to other contexts.
[26]. For example, in the field of causal ML, real-life field tests are a Furthermore, the choice of evaluation metrics plays a pivotal role in
common tool for generating experimental data to check the beneficial aligning AI solutions with the objectives of decision-makers. Histori­
impact of different treatments on the final outcome. For example, [10] cally, the IS literature has relied on statistical metrics such as accuracy,
tests the impact of three different motivational emails sent to members recall, precision, area under the ROC curve (AUC), and H-measure for
of the online community to measure whether motivational emails have predictive model evaluation. However, these metrics often lack direct
an impact on member participation and answer the question of which interpretability and may not correlate with standard business metrics
motivational email works best for which member. [27] rely on empirical and key performance indicators. Hence, a recent surge in research has
data gathered through experiments, emphasizing the importance of data emerged within the field of profit analytics, focusing on optimizing and
quality for the validity of their serial mediation model. [28] particularly evaluating algorithms based on profit-oriented criteria. For example,
focusing on decision-making within sharing economy platforms in the [38] introduced ProfLogit, an adaptation of conventional logistic
Metaverse. It employs a case study to empirically validate its theoretical regression designed to maximize the expected maximum profit criterion
propositions, contributing to both the academic and practical under­ (EMPC). In the context of churn prediction, their work demonstrated
standing of the subject. The study contributes to the data aspect of XAI that ProfLogit outperforms other models, achieving the highest out-of-
by employing fuzzy logic, specifically through the use of Schweizer- sample EMPC performance, as well as the best profit-based precision
Sklar norms. This approach is designed to manage uncertainty in deci­ and recall values overall. In the field of marketing, [39] developed
sion making within the Metaverse sharing economy. Although it does profit-conscious ensemble selection (PCES), demonstrating that the
not revolutionize the field of XAI, it does offer a way to handle ambig­ resulting measures offer a managerially meaningful assessment of the
uous or uncertain data in a specific application context. The synergy business value of a targeting model and the extent to which PCES im­
between XAI and these types of data is essential to build trust in AI proves decision quality given that profit is a more interpretable business
systems, address bias and fairness concerns, and ensure that AI-based metrics compared to statistical metrics such as AUC or accuracy.
decisions are not only accurate but also explainable, interpretable, and
actionable for stakeholders and end users. It also facilitates regulatory 2.3. Application
compliance and ethical considerations in AI applications in various
domains. The existing literature in the field of IS explores various decision-
making scenarios where AI plays a central role. First, it sheds light on
2.2. Method mature IS applications that have long benefited from ML advancements
but have more recently incorporated emerging trends in the field of XAI.
ML-based AI solutions require careful consideration when selecting Notable application domains include credit scoring [40], customer
the appropriate algorithms. In the realm of extant IS literature, there is a churn prediction [31], educational science [41], identifying factors
fundamental distinction between”glass-box” and”black-box” algo­ influencing accident severity [42] and fraud detection [43]. For
rithms. Glassbox algorithms offer direct interpretability, providing de­ example, [31] introduces a novel ML algorithm, the spline-rule
cision makers with insights into the decision-making process. In ensemble classifier with structured sparsity regularization, offering an
particular, regression models and decision trees are prevalent examples effective balance between prediction accuracy and strong interpret­
of glass-box models found in the XAI literature. For example, [3] used a ability. [44] explores the use of probabilistic ML to understand moti­
logistic regression model to reveal the key student metrics that influence vational climate variables and their impact on student teaching
dropout behavior. They leveraged the beta coefficients of their logistic evaluations, improving decision making through XAI. [42] introduces
regression model to visually illustrate the direction and impact of each an explainable analytics framework combining descriptive, predictive,
metric on student attrition. In addition, the advent of Generalized Ad­ and prescriptive analytics to enhance the transparency and decision-
ditive Models (GAMs) has introduced a nuanced perspective that aims to making process to understand factors that contribute to the severity of
bridge the gap between model interpretability and predictive perfor­ accidents.
mance [29]. Particularly capable of handling structured data, GAMs and Conversely, we observe a growing interest in IS literature towards
related spline-base methods have been applied to various prediction developing XAI solutions in domains that are not predominantly data-
tasks, such as house price prediction [30] or customer churn prediction driven and ML-oriented. These newer applications encompass areas
[31]. such as student attrition prediction [3] and human resource manage­
However, a prevailing assumption exists that black-box models, ment [45]. For example, [3] visualize the course feedback given by the
owing to their potentially superior predictive performance, lead to more students using t-SNE. The authors find two segments for students that
informed decisions than their glass-box counterparts. Recent discuss various topics, including feedback related to the teacher, the

2
K. Coussement et al. Decision Support Systems 184 (2024) 114276

course, the facilities, and student issues. In the context of a health crisis elucidates these effects, providing a clear analytical framework.
like a pandemic, the researchers do not just stop at developing ML Empirical tests confirm the model’s efficacy, and its interpretability
models. They take it a step further by integrating it into a real-world offers significant insights for managerial decision-making.
setting. Using a blend of advanced analytics [46] and Bayesian net­ The persistent pilot shortage in the USAF is being worsened by the
works [47], they create a decision-making tool that healthcare pro­ rising demand from commercial airlines. To tackle this problem, [52]
fessionals can actually use. [28] addresses the practical application of contribute to the IS literature by proposing a decision support system for
XAI by exploring the sharing economy within the emerging field of the selecting pilot candidates using advanced ML techniques. Considering
Metaverse. The study moves beyond theoretical propositions by the recent Responsible Artificial Intelligence Strategy released by the US
including a case study for empirical validation. While the case study is Department of Defense, this study employs interpretable and explain­
limited in scope, it provides a useful starting point for further research able ML methods to develop traceable and fair models that can be
and offers some actionable insights for stakeholders in this specific area. governed responsibly and reliably. These models are utilized to predict
candidates’ average merit assignment selection system scores based on
3. Overview of contributions to this special issue pre-selection and pre-training information.
The following paper addresses the lag in AI adoption among industry
This section summarizes the papers of the special issue and gives players by proposing XAI as a solution to improve decision-making
insight into how they contribute to enhanced decision-making through processes. Using an experimental design, [27] provides empirical evi­
XAI. dence on the impact of XAI on supply chain decision making. The study
Online retail platforms face numerous challenges, such as cyber- identifies a serial mediation path that involves transparency and agile
attacks, data breaches, and device failures, that jeopardize their opera­ decision making, showing that XAI enhances transparency and improves
tional integrity and data security. [48] propose a robust resilience cyber resilience during supply chain cyberattacks. A post-hoc text
strategy that integrates explainable deep learning with a blockchain- analysis of X posts reveals a positive attitude towards XAI and highlights
based consensus protocol to address these threats. By leveraging this themes of transparency, explainability, and interpretability. In general,
combined approach, the strategy facilitates rapid detection of incidents, the findings underscore the significant role of XAI in improving decision
provides clear explanations of detected vulnerabilities, and improves support systems.
decision-making processes. To validate the strategy’s effectiveness, the [53] introduce a comprehensive framework for expert-augmented
authors conducted an experimental study using NAB datasets on real- supervised feature selection, covering pre-processing, in-processing,
time online retail systems. The results highlight the strategy’s ability and post-processing aspects of XAI. In the pre-processing stage, the
to implement efficient and effective cyber resilience measures, thereby Probabilistic Solution Generator through Information Fusion (PSGIF)
enhancing the decision-making capabilities necessary for maintaining algorithm is introduced, using ensemble techniques to enhance a Ge­
business continuity and operational resilience. netic Algorithm’s (GA) capabilities. The framework balances explain­
[28] contextualize their study in the intersection of the sharing ability and prediction accuracy with multiobjective optimization
economy and the Metaverse. Transportation applications in the sharing models, allowing experts to set acceptable sacrifice levels, thus
economy reduce car ownership and single-vehicle use, enhancing improving explainability by focusing on relevant features. The in-
regional environmental sustainability. The Metaverse integrates these processing stage incorporates expert opinions into feature selection as
applications with transportation networks, bolstering sustainability. a multi-objective problem, optimizing GA parameters through Bayesian
This research supports decision makers by creating an explainable multi- optimization. For post-processing, the Posterior Ensemble Algorithm
criterion decision-making (MCDM) model to prioritize three Metaverse (PEA) assesses the predictive power of features, comparing objective
integration alternatives: safety measures, payment systems, and opti­ and subjective importance, and is tested on 16 datasets to validate the
mizing operations. The authors showcase a case study with twelve framework’s effectiveness in both single and multi-objective settings.
criteria in four aspects (economic, user, operational, and advancement) In the e-commerce sector, [54] explores how textual product de­
provides expert evaluations. Findings suggest optimizing operations scriptions affect consumer choices and sales forecasting. The authors
through Metaverse integration is the best approach. create an AI framework combining text mining, advanced neural net­
In this study, [49] introduce MERLIN, an innovative XAI method that works, and regression analysis. By analyzing nearly 200,000 sales re­
offers contrastive explanations for two ML models, thereby presenting cords, they demonstrate the use of SHAP to uncover the significance of
the idea of model-contrastive explanations. The researchers develop an specific product description phrases in enhancing sales forecasting and
encoding that enables MERLIN to handle both textual and tabular data, customer acquisition. This study presents a transparent and under­
as well as a mix of continuous and discrete features. To demonstrate the standable model that uses textual data to improve e-commerce
efficacy of their method, they test MERLIN on a wide range of bench­ strategies.
mark datasets. Additionally, the authors enhance the IS field by Complying with regulatory frameworks such as GDPR becomes a
releasing MERLIN as a publicly accessible python-pip package. crucial point when developing AI systems in high-stakes domains. Often,
[50] present a hybrid XAI framework to evaluate cyber-risks arising black-box models are not an option in these scenarios. [30] delves into
from correlated phishing attacks. Their framework is divided into this topic by introducing LitBoost, a tree-based boosting model that
several phases. The initial phase determines the probability of expert improves the interpretability of gradient boosted trees by modifying
phishers within a community of attackers with varying levels of exper­ them to work similarly to GAMs, known for their transparency and
tise. The subsequent phase assesses the likelihood of phishing attacks on simplicity. The paper evaluates LitBoost using a dataset from the Oslo
a firm despite its investments in IT security and regulatory measures. housing market and a synthetic dataset designed to mimic real world
The third phase differentiates phishing and legitimate URLs using heterogeneity. The results demonstrate LitBoost’s ability to provide
various ML classifiers. Following this, it estimates the joint distribution accurate and interpretable predictions, while enabling information
of phishing attacks using an exponential-beta distribution and measures sharing between different geographic groups during the training
the expected loss with Archimedean Copula. Finally, the authors provide process.
recommendations for firms by calculating optimal investments in cyber- [55] introduces a decision support system tailored for proactive
insurance versus IT security. management in railway control rooms. This system offers explainable
[51] introduce an innovative and interpretable Graph Neural predictions and recommendations on the automation usage in traffic
Network (GNN) model designed for assessing financial distress. This control, where operators make critical decisions on signal operations.
model incorporates financial indicators, network data, and event in­ Emphasizing the need for explainability in automation, particularly in
formation. It not only identifies the propagation effects of events but also settings where split-second decisions are crucial, this decision support

3
K. Coussement et al. Decision Support Systems 184 (2024) 114276

system employs SHAP values to ensure consistency in explainability decision-making.

across various ML methods. Implemented as a proof of concept in These papers collectively showcase innovative methodologies that
Infrabel’s digital control rooms, the system’s effectiveness and impact significantly enhance decision-making through the application of XAI,
on operational management were validated through end-user feedback, addressing current challenges and proposing advanced solutions to
focusing on the system’s operational impact and the perceived value of foster future advancements in the field.
incorporating agreement levels in explainability.
Shapley values are one of the main tools for model-agnostic post hoc 4. Final notes & future research agenda
explanations on tabular datasets. In [56], the authors propose a novel
approach that combines two different explanations provided by Shapley This special issue shows that XAI is an important stream of research
values; one that explains the model itself and another that explains the to further enhance decision making along the data, method and appli­
model combined with feature dependencies in the data. This approach cation side. Furthermore, various future research directions emerge that
eliminates the need to choose between the two explanations and en­ highlight several promising research areas. For example, within the area
hances the overall explanatory power of Shapley values, offering intui­ of e-commerce, future studies could investigate whether the relatively
tive and superior insights on real-world datasets. low impact of product descriptions on sales, compared to historical sales
Outlier detection is a very challenging application for XAI techniques data and prices, is due to the intrinsic nature of the data or to the loss of
because it typically relies on unsupervised learning. In [57], the authors information during text processing [54]. Especially with recent ad­
address the task of data categorization by building on formal concept vances in natural language processing, this line of research can be of
analysis (FCA). They recognize that different sets of features used for great interest to a wide IS readership. In the context of human-machine
classification can yield varying categorizations, influenced by subjective interaction, future research could study the addition of detailed operator
choices based on interests or preferences, referred to as interrogative information to the model, which is expected to greatly impact the pre­
agendas. They represent these agendas as sets of features and develop diction accuracy for future human operator behavior [55]. In the area of
two algorithms: an unsupervised FCA-based algorithm for outlier methodological innovations, more research could explore the integra­
detection using different agendas and a supervised meta-learning algo­ tion of Structured Additive Regression (STAR) models with the idea of
rithm to learn suitable (fuzzy) agendas with different feature weights. training local models to improve predictive accuracy while maintaining
They combine these to create a supervised outlier detection algorithm, interpretability [30]. Furthermore, one of the main challenges in the
showing that their approach performs comparably to standard outlier current XAI landscape is to improve the interpretation of multimodal
detection algorithms on benchmark datasets and provides both local and models, which involve integrating multiple types of data (e.g., text,
global explanations of the results. images, audio, and structured data). By combining diverse data sources,
Another approach based on Shapley values was proposed in [58]. multimodal XAI can provide richer and more comprehensive explana­
The authors focus on improving customer churn prediction by utilizing tions of the prediction of the model. In this special issue, several
hybrid models that apply model segmentation and classification. The methodological advances have been presented mainly for tabular data
goal of these approaches is to capture non-linearities while ensuring (see, e.g., [56–58]), textual data [49], and network data [51], suggesting
interpretability. The authors introduce a novel model-agnostic approach that explainability is a key element for unimodal decision making.
based on SHAP to interpret these complex models. The proposed method However, there is a pressing need for multimodal methods that can
is extensively benchmarked on 14 customer churn datasets for predic­ derive interpretable insights from multimodal tasks [62]. For example,
tive performance, and the interpretability is demonstrated through a [57,58] show active research on XAI methods based on Shapley values,
case study. The contributions include introducing new hybrid especially for business decision support systems. These methods can be
segmented models and proposing a modelagnostic tool for interpreting further extended and combined with other XAI strategies for deep
hybrid segmented models, relevant for decision makers in various pre­ learning to provide integrated solutions for multimodal systems.
dictive modeling tasks beyond customer retention management. Building on the work of [59–61] future studies should explore the
[59] introduce an innovative method utilizing Representational application of RSA to broader AI ecosystems, particularly focusing on
Similarity Analysis (RSA) to evaluate explanation similarity within XAI the long-term impacts of explanation stability on user trust and system
applications. The study demonstrates that RSA can quantitatively assess adoption. In addition, future research should investigate the scalability
the stability of explanations across various ML models and preprocessing of advanced predictive clustering models and their application to
techniques. By mitigating explanation drift, RSA enhances the reliability diverse datasets. Enhancing these models with additional layers of
and consistency of XAI systems, thereby supporting more robust explainability, such as incorporating domain-specific knowledge, could
decision-making frameworks. The findings indicate that RSA signifi­ further improve their interpretability and utility. In addition, refining
cantly improves the interpretability of ML models, leading to higher counterfactual explanation methods to handle more complex, multi-
trust and better-informed decisions. dimensional datasets and improve their applicability across various
[60] propose an advanced predictive clustering model integrated business domains. Future work could also focus on developing more
with the Dirichlet Process Mixture Model (DPMM) to handle heteroge­ personalized counterfactual explanations to meet specific user needs,
neous and longitudinal electronic health records (EHRs). By incorpo­ enhancing decision making in personalized marketing strategies [61].
rating attention mechanisms, the model not only improves disease risk
prediction but also offers multi-level explainable evidence. Evaluations
on real-world datasets confirm the model’s effectiveness in capturing Declaration of competing interest
detailed patient trajectories and providing actionable insights for
healthcare practitioners. Performance evaluations reveal that the model No potential conflict of interest was reported by the authors.
achieves a higher accuracy and interpretability compared to traditional
clustering methods.
[61] presents a methodology for generating counterfactual expla­ Acknowledgments
nations that recommend business changes to improve customer
repurchase behavior. By utilizing electronic word of mouth (eWOM) The guest editorial team thanks Prof. Victoria Yoon and Prof. Andrew
data, the method predicts customer revisit intentions and provides Chen for their trust in this project and their leadership in guiding this
actionable insights for different customer segments. The empirical special issue. Furthermore, we thank the authors of the papers collected
evaluation demonstrates the method’s superior performance compared in this special issue for their professionalism in incorporating the Re­
to existing approaches, making it a valuable tool for enhancing business viewers’ and Editors’ comments and suggestions.

4
K. Coussement et al. Decision Support Systems 184 (2024) 114276

References [27] K. Sadeghi R, D. Ojha, P. Kaur, R.V. Mahto, A. Dhir, Explainable artificial intelli­
gence and agile decision-making in supply chain cyber resilience, Decis. Support.
Syst. 180 (2024) 114194, [Link]
[1] K. Topuz, F.D. Zengul, A. Dag, A. Almehmi, M.B. Yildirim, Predicting graft survival
[28] D. Pamucar, M. Deveci, I. Gokasar, D. Delen, M. Koppen, W. Pedrycz, Evaluation of
among kidney transplant recipients: a Bayesian decision support model, Decis.
metaverse integration alternatives of sharing economy in transportation using
Support. Syst. 106 (2018) 97–109, [Link]
fuzzy Schweizer-Sklar based ordinal priority approach, Decis. Support. Syst. 171
[2] M. Kraus, S. Feuerriegel, A. Oztekin, Deep learning in business analytics and op­
(2023) 113944, [Link]
erations research: models, applications and managerial implications, Eur. J. Oper.
[29] M. Kraus, D. Tschernutter, S. Weinzierl, P. Zschech, Interpretable generalized ad­
Res. 281 (3) (2020) 628–641, [Link]
ditive neural networks, Eur. J. Oper. Res. 317 (2) (2024).
[3] M. Phan, A. De Caigny, K. Coussement, A decision support framework to incor­
[30] A. Hjort, I. Scheel, D.E. Sommervoll, J. Pensar, Locally interpretable tree boosting:
porate textual data for early student dropout prediction in higher education, Decis.
an application to house price prediction, Decis. Support. Syst. 178 (2024) 114106,
Support. Syst. 168 (January) (2023) 113940, [Link]
[Link]
dss.2023.113940.
[31] K.W. De Bock, A. De Caigny, Spline-rule ensemble classifiers with structured
[4] E. Eryarsoy, K. Topuz, C. Demiroglu, Disentangling human trafficking types and the
sparsity regularization for interpretable customer churn modeling, Decis. Support.
identification of pathways to forced labor and sex: an explainable analytics
Syst. 150 (2021) 113523 interpretable Data Science For Decision Making, htt
approach, Ann. Oper. Res. (2023) 1–35.
ps://[Link]/10.1016/[Link].2021.113523.
[5] F. Vandervorst, W. Verbeke, T. Verdonck, Data misrepresentation detection for
[32] Z. Yi, Z. Liang, T. Xie, F. Li, Financial risk prediction in supply chain finance based
insurance underwriting fraud prevention, Decis. Support. Syst. 159 (2022) 113798,
on buyer transaction behavior, Decis. Support. Syst. 170 (2023) 113964, https://
[Link]
[Link]/10.1016/[Link].2023.113964.
[6] R. Fu, Y. Huang, P.V. Singh, Crowds, lending, machine, and bias, Inf. Syst. Res. 32
[33] Y. Huang, S. Meng, Automobile insurance classification ratemaking based on tel­
(1) (2021) 72–92, [Link]
ematics driving data, Decis. Support. Syst. 127 (2019) 113156.
[7] A. Fugener, J. Grahl, A. Gupta, W. Ketter, Cognitive challenges in human–artificial
[34] S. Lundberg, S.-I. Lee (Eds.), Proceedings of the 31st International Conference on
intelligence collaboration: investigating the path toward productive delegation,
Neural Information Processing Systems, NIPS’17, 2017.
Inf. Syst. Res. (2022), [Link]
[35] A. Goldstein, A. Kapelner, J. Bleich, E. Pitkin, Peeking inside the black box: visu­
[8] S. Fan, N. Ilk, A. Kumar, R. Xu, J.L. Zhao, Blockchain as a trust machine: from
alizing statistical learning with plots of individual conditional expectation,
disillusionment to enlightenment in the era of generative AI, Decis. Support. Syst.
J. Comput. Graph. Stat. 24 (1) (2015) 44–65.
182 (2024) 114251, [Link]
[36] M.T. Ribeiro, S. Singh, C. Guestrin (Eds.), Proceedings of the 22nd ACM SIGKDD
[9] K. Idbenjra, K. Coussement, A. De Caigny, Investigating the beneficial impact of
International Conference on Knowledge Discovery and Data Mining, ACM, 2016.
segmentation-based modelling for credit scoring, Decis. Support. Syst. 179 (2024)
[37] B. Davazdahemami, H.M. Zolbanin, D. Delen, An explanatory machine learning
114170, [Link]
framework for studying pandemics: the case of covid-19 emergency department
[10] S. Debaere, F. Devriendt, J. Brunneder, W. Verbeke, T. De Ruyck, K. Coussement,
readmissions, Decis. Support. Syst. 161 (2022) 113730 data Analytics and
Reducing inferior member community participation using uplift modeling: evi­
Decision-Making Systems: Implications of the Global Outbreaks, [Link]
dence from a field experiment, Decis. Support. Syst. 123 (August) (2019), https://
org/10.1016/[Link].2022.113730.
[Link]/10.1016/[Link].2019.113077.
[38] E. Stripling, S. vanden Broucke, K. Antonio, B. Baesens, M. Snoeck, Profit maxi­
[11] A. De Caigny, K. Coussement, K.W. De Bock, Leveraging fine-grained transaction
mizing logistic model for customer churn prediction using genetic algorithms,
data for customer life event predictions, Decis. Support. Syst. 130 (May 2019)
Swarm Evol. Comput. 40 (September) (2018) 116–130, [Link]
(2020) 113232, [Link]
swevo.2017.10.010.
[12] K. Coussement, S. Lessmann, G. Verstraeten, A comparative analysis of data
[39] S. Lessmann, J. Haupt, K. Coussement, K.W. De Bock, Targeting customers for
preparation algorithms for customer churn prediction: a case study in the tele­
profit: an ensemble learning framework to support marketing decision-making, Inf.
communication industry, Decis. Support. Syst. 95 (2017) 27–36, [Link]
Sci. 557 (2021) 286–301.
10.1016/[Link].2016.11.007.
[40] Z. Wang, J. Xiao, L. Wang, J. Yao, A novel federated learning approach with
[13] Y. Jung, Y. Suh, Mining the voice of employees: a text mining approach to iden­
knowledge transfer for credit scoring, Decis. Support. Syst. (2023) 114084, https://
tifying and analyzing job satisfaction factors from online employee reviews, Decis.
[Link]/10.1016/[Link].2023.114084.
Support. Syst. 123 (2019) 113074, [Link]
[41] K. Topuz, B.D. Jones, S. Sahbaz, M. Moqbel, Methodology to combine theoretical
[14] V.-H. Trieu, H.Q. Vu, M. Indulska, G. Li, A computer vision-based concept model to
knowledge with a data-driven probabilistic graphical model, J. Bus. Anal. 4 (2)
recommend domestic overseas-like travel experiences: a design science study,
(2021) 125–139.
Decis. Support. Syst. 181 (2024) 114149, [Link]
[42] A. Ahmed, K. Topuz, M. Moqbel, I. Abdulrashid, What makes accidents severe!
dss.2023.114149.
Explainable analytics framework with parameter optimization, Eur. J. Oper. Res.
[15] B. Cankaya, K. Topuz, D. Delen, A. Glassman, Evidence-based managerial decision-
317 (2) (2024).
making with machine learning: the case of bayesian inference in aviation incidents,
[43] R. Van Belle, B. Baesens, J. De Weerdt, Catchm: a novel network-based credit card
Omega 102906 (2023).
fraud detection method using node representation learning, Decis. Support. Syst.
[16] H. Liu, Y. Ye, H.Y. Lee, High-dimensional learning under approximate sparsity with
164 (2023) 113866, [Link]
applications to nonsmooth estimation and regularized neural networks, Oper. Res.
[44] B.D. Jones, K. Topuz, S. Sahbaz, Predicting undergraduate student evaluations of
(2022), [Link]
teaching using probabilistic machine learning: the importance of motivational
[17] P. Borchert, K. Coussement, A. De Caigny, J. De Weerdt, Extending business failure
climate, Stud. Educ. Eval. 81 (2024) 101353, [Link]
prediction models with textual website content using deep learning, Eur. J. Oper.
stueduc.2024.101353.
Res. 306 (1) (2022) 348–357, [Link]
[45] N. Lawrance, G. Petrides, M.-A. Guerry, Predicting employee absenteeism for cost
[18] D. Shin, The effects of explainability and causability on perception, trust, and
effective interventions, Decis. Support. Syst. 147 (2021) 113539, [Link]
acceptance: implications for explainable AI, Int. J. Hum.-Comput. Stud. 146 (2021)
10.1016/[Link].2021.113539.
102551, [Link]
[46] K. Topuz, D. Delen, A probabilistic bayesian inference model to investigate injury
[19] A. Rai, Explainable AI: from black box to glass box, J. Acad. Mark. Sci. 48 (1)
severity in automobile crashes, Decis. Support. Syst. 150 (2021) 113557.
(2020) 137–141, [Link]
[47] K. Topuz, B. Davazdahemami, D. Delen, A bayesian belief networkbased analytics
[20] B. Kim, J. Park, J. Suh, Transparency and accountability in AI decision support:
methodology for early-stage risk detection of novel diseases, Ann. Oper. Res.
explaining and visualizing convolutional neural networks for text information,
(2023) 1–25.
Decis. Support. Syst. 134 (July 2019) (2020) 113302, [Link]
[48] K. Zkik, A. Belhadi, S. Kamble, M. Venkatesh, O. Mustapha, A. Sebbar, Cyber
dss.2020.113302.
resilience framework for online retail using explainable deep learning approaches
[21] K. Topuz, T.L. Urban, M.B. Yildirim, A markovian score model for evaluating
and blockchainbased consensus protocol, Decis. Support. Syst. (2024) 114253,
provider performance for continuity of care—an explainable analytics approach,
[Link]
Eur. J. Oper. Res. 317 (2) (2024).
[49] L. Malandri, F. Mercorio, M. Mezzanzanica, A. Seveso, Model-contrastive expla­
[22] D. Zhdanov, S. Bhattacharjee, M.A. Bragin, Incorporating FAT and privacy aware
nations through symbolic reasoning, Decis. Support. Syst. 176 (2024) 114040,
AI modeling approaches into business decision making frameworks, Decis. Sup­
[Link]
port. Syst. 155 (2022) 113715, [Link]
[50] B. Biswas, A. Mukhopadhyay, A. Kumar, D. Delen, A hybrid framework using
[23] G. Phillips-Wren, M. Daly, F. Burstein, Reconciling business intelligence, analytics
explainable AI (XAI) in cyber-risk management for defence and recovery against
and decision support systems: more data, deeper insight, Decis. Support. Syst. 146
phishing attacks, Decis. Support. Syst. 177 (2024) 114102, [Link]
(March) (2021) 113560, [Link]
10.1016/[Link].2023.114102.
[24] L. Wang, R. Gopal, R. Shankar, J. Pancras, Forecasting venue popularity on
[51] J. Wang, C. Jiang, L. Zhou, Z. Wang, Assessing financial distress of SMEs through
location-based services using interpretable machine learning, Prod. Oper. Manag.
event propagation: an adaptive interpretable graph contrastive learning model,
(2022), [Link]
Decis. Support. Syst. 180 (2024) 114195, [Link]
[25] P. Harrigan, T.M. Daly, K. Coussement, J.J.A.J. Lee, G.G.N. Soutar, U. Evers,
dss.2024.114195.
Identifying influencers on social media, Int. J. Inf. Manag. 56 (September 2019)
[52] D. Wasilefsky, W.N. Caballero, C. Johnstone, N. Gaw, P.R. Jenkins, Responsible
(2021) 102246, [Link]
machine learning for United States Air Force pilot candidate selection, Decis.
[26] A. De Caigny, K. Coussement, W. Verbeke, K. Idbenjra, M. Phan, Uplift modeling
Support. Syst. 180 (2024) 114198, [Link]
and its implications for B2B customer churn prediction: a segmentation-based
[53] M. Rabiee, M. Mirhashemi, M.S. Pangburn, S. Piri, D. Delen, Towards explainable
modeling approach, Ind. Mark. Manag. 99 (September) (2021) 28–39, [Link]
artificial intelligence through expert-augmented supervised feature selection,
org/10.1016/[Link].2021.10.001.
Decis. Support. Syst. 181 (2024) 114214, [Link]
dss.2024.114214.

5
K. Coussement et al. Decision Support Systems 184 (2024) 114276

[54] S. Chen, S. Ke, S. Han, S. Gupta, U. Sivarajah, Which product description phrases comprehensive review of methodological advances and future research directions,
affect sales forecasting? An explainable AI framework by integrating WaveNet arXiv preprint arXiv:2306.05731, 2023.
neural network models with multiple regression, Decis. Support. Syst. 176 (2024)
114065, [Link]
[55] L. Sobrie, M. Verschelde, Real-time decision support for human–machine interac­ Kristof Coussementa,1,*, Mohammad Zoynul Abedinb,1,
tion in digital railway control rooms, Decis. Support. Syst. 181 (2024) 114216, Mathias Krausc,1, Sebastián Maldonadod,f,1, Kazim Topuze,1
[Link] a
IESEG School of Management Univ. Lille CNRS UMR 9221, LEM, Lille
[56] J. Michiels, J. Suykens, M. De Vos, Explaining the model and feature dependencies
by decomposition of the Shapley value, Decis. Support. Syst. 182 (2024) 114234,
Economie Management, 3 Rue de la Digue, 59000 Lille, France
b
[Link] Swansea University, Department of Accounting and Finance, School of
[57] M. Boersma, K. Manoorkar, A. Palmigiano, M. Panettiere, A. Tzimoulis, Management Bay Campus, Fabian Way, Swansea SA1 8EN, Wales, United
N. Wijnberg, Outlier detection using flexible categorization and interrogative
agendas, Decis. Support. Syst. 180 (2024) 114196, [Link]
Kingdom
c
dss.2024.114196. University of Regensburg, Faculty for Informatics and Data Science,
[58] A. De Caigny, K.W. De Bock, S. Verboven, Hybrid black-box classification for Bajuwarenstrae 4, 93053 Regensburg, Germany
customer churn prediction with segmented interpretability analysis, Decis. Sup­ d
University of Chile, Department of Management Control and Information
port. Syst. 181 (2024) 114217, [Link]
[59] F. Gwinner, C. Tomitza, A. Winkelmann, Comparing expert systems and their Systems, School of Economics and Business, Santiago, Chile
e
explainability through similarity, Decis. Support. Syst. (2024) 114248, [Link] The University of Tulsa, Collins College of Business, School of Finance and
org/10.1016/[Link].2024.114248. Operations Management, Tulsa, OK, USA
[60] A. Artelt, A. Gregoriades, Supporting organizational decisions on how to improve f
customer repurchase using multi-instance counterfactual explanations, Decis. Instituto Sistemas Complejos de Ingeniería (ISCI), Santiago, Chile
Support. Syst. (2024) 114249, [Link]
[61] S. Niu, Q. Yin, J. Ma, Y. Song, Y. Xu, L. Bai, W. Pan, X. Yang, Enhancing healthcare *
decision support through explainable AI models for risk prediction, Decis. Support.
Corresponding author.
Syst. 181 (2024) 114228, [Link] E-mail address: [Link]@[Link] (K. Coussement).
[62] N. Rodis, C. Sardianos, G.T. Papadopoulos, P. Radoglou-Grammatikis,
P. Sarigiannidis, I. Varlamis, Multimodal explainable artificial intelligence: A

1
All authors did contribute equally.