Explainable Artificial Intelligence

Chronic kidney disease (CKD) is a major worldwide health problem, affecting a large proportion of the world's population and leading to higher morbidity and death rates. The early stages of CKD sometimes present without visible symptoms, causing patients to be unaware. Early detection and treatments are critical in reducing complications and improving the overall quality of life for people afflicted. In this work, we investigate the use of an explainable artificial intelligence (XAI)-based strategy, leveraging clinical characteristics, to predict CKD. This study collected clinical data from 491 patients, comprising 56 with CKD and 435 without CKD, encompassing clinical, laboratory, and demographic variables. To develop the predictive model, five machine learning (ML) methods, namely logistic regression (LR), random forest (RF), decision tree (DT), Naïve Bayes (NB), and extreme gradient boosting (XGBoost), were employed. The optimal model was selected based on accuracy and area under the curve (AUC). Additionally, the SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) algorithms were utilized to demonstrate the influence of the features on the optimal model. Among the five models developed, the XGBoost model achieved the best performance with an AUC of 0.9689 and an accuracy of 93.29%. The analysis of feature importance revealed that creatinine, glycosylated hemoglobin type A1C (HgbA1C), and age were the three most influential features in the XGBoost model. The SHAP force analysis further illustrated the model's visualization of individualized CKD predictions. For further insights into individual predictions, we also utilized the LIME algorithm. This study presents an interpretable ML-based approach for the early prediction of CKD. The SHAP and LIME methods enhance the interpretability of ML models and help clinicians better understand the rationale behind the predicted outcomes more effectively. Chronic kidney disease (CKD) is a progressive and often debilitating medical condition characterized by the gradual loss of kidney function over time. It is defined by a glomerular filtration rate (GFR) under 60 mL/min/1.73 m 2 for over 3 months or the presence of kidney damage indicators 1,2 . Common causes of CKD include diabetes, high blood pressure, glomerulonephritis, and other related conditions 3 . It affects millions globally and creates a significant healthcare burden due to its high prevalence, costly treatments, and potentially life-threatening complications 4 . Early detection and identification of critical risk factors are crucial to managing CKD effectively and improving patient outcomes. Traditional diagnostic approaches are ineffective in properly predicting the progression of CKD and identifying the most significant factors contributing to its development. However, recent advancements in computational intelligence and machine learning (ML) techniques have shown great promise in terms of revolutionizing CKD diagnosis and variable importance analysis . This introduction investigates the relationship between computational intelligence and CKD diagnosis, emphasizing the use of ML techniques to predict and diagnose the disease at early stages. It also explores the idea of variable importance, highlighting its significance in understanding the underlying factors driving CKD progression and the development of tailored treatment strategies. Over recent years, various studies have focused on the effective and precise diagnosis of CKD patients. Diverse classifiers have been employed to establish classification models to categorize clinical CKD data, as evidenced by the work of Jena et al. . The investigation carried out by Ahmad et al. 10 utilized supervised ML techniques, including neural networks (NN), logistic regression (LR), and support vector machines (SVM), to achieve this objective. A unified approach combining classification and association rule mining techniques has been employed to design a system for predicting and diagnosing CKD and its underlying causes. Among the diverse algorithms explored on medical data are random forest (RF), k-nearest neighbors (KNN), decision trees (DT), Naïve Bayes

Thyroid disease classification is a critical challenge in medical diagnostics, requiring accurate differentiation between hyperthyroidism, hypothyroidism, and normal thyroid function. This study introduces an advanced machine learning approach that integrates generative adversarial networks (GANs) for data augmentation and Kolmogorov-Arnold networks (KANs) for classification. Various machine learning models including logistic regression, random forest, support vector machines, multilayer perceptrons, and KANs were trained and evaluated. The results indicate that the application of GAN-based data augmentation has significantly improved classification accuracy, particularly for minority classes. Specifically, the KAN model achieved an accuracy of 98.68% and random forest (RF) F1-score of 98.00%, outperforming traditional neural network applications. The results demonstrate that GAN-augmented datasets significantly improve classification accuracy, and the KAN model achieves superior performance and generalization capabilities compared to traditional neural networks. Additionally, the SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) were employed to ensure model transparency and interpretability. These explainability methods highlight thyroid stimulating hormone as the most prominent feature in classification, further supporting its clinical utility in the diagnosis of thyroid diseases. The findings underscore the potential of advanced AI-driven techniques in improving thyroid disease classification, addressing class imbalance, and enhancing explainability in healthcare applications. By leveraging synthetic data generation, this study provides a feasible framework for actual clinical application, particularly in situations where clinical data are limited or imbalanced. The integration of GANs and KANs enhances diagnostic accuracy while preserving robustness and generalizability to diverse patient populations. Besides, the approach fosters the deployment of explainable AI models in clinical decision support systems so that healthcare practitioners can make improved and more reliable decisions, thus leading to better patient outcomes and resource allocation.

Federated Learning (FL) provides both model performance and data privacy for machine learning tasks where samples or features are distributed among different parties. In the training process of FL, no party has a global view of data distributions or model architectures of other parties. Thus the manually-designed architectures may not be optimal. In the past, Neural Architecture Search (NAS) has been applied to FL to address this critical issue. However, existing Federated NAS approaches require prohibitive communication and computation effort, as well as the availability of high-quality labels. In this work, we present Self-supervised Vertical Federated Neural Architecture Search (SS-VFNAS) for automating FL where participants hold feature-partitioned data, a common cross-silo scenario called Vertical Federated Learning (VFL). In the proposed framework, each party first conducts NAS using self-supervised approach to find a local optimal architecture with its own data. Then, parties...

The Common Vulnerabilities and Exposures (CVE) database is the largest publicly available source of structured data on software and hardware vulnerability. In this work, we analyze the CVE database in the context of IoT device and system vulnerabilities. We employ and compare support vector machine (SVM) and neural network (NN) algorithms on a selected subset of the CVE database to classify vulnerability records in this framework. Our scope of interest consists of records that describe vulnerabilities of potential IoT devices of different types, such as home appliances, SCADA (industry) devices, mobile controllers, networking equipment and others. The purpose of this work is to develop and test an automated system of recognition of IoT vulnerabilities to test two different methods of classification (SVM and NN) and to find an optimal timeframe for training (historical) data.

Artificial intelligence (AI) has achieved unprecedented advancements in recent years, driven largely by deep learning architectures and large-scale data analytics (LeCun, Bengio, & Hinton, 2015). Despite these achievements, there remains a critical gap between computational efficiency and the adaptive, context-aware reasoning characteristic of human cognition (Marcus, 2020). In this study, we put forward Hybrid Cognitive–Machine Learning (HCML), a novel paradigm integrating cognitive models of human reasoning with state-of-the-art machine learning techniques. The framework is designed to enhance interpretability, ethical alignment, and generalization in AI systems. Drawing from high-impact literature, HCML proposes a bidirectional intelligence exchange model, enabling co-learning between humans and machines. Applications in healthcare, education, climate science, and policy-making are discussed, supported by a comprehensive literature review and methodological blueprint. The proposed paradigm holds the potential to reshape both academic research and industrial practice, providing a foundation for more trustworthy and human-aligned AI systems.

In the competitive world of digital banking, predicting and reducing customer churn is essential for long-term growth. Traditional predictive models can forecast churn quite accurately, but their lack of transparency is a problem in regulated areas like finance, where clarity and responsibility are crucial. This study looks into how to combine Explainable Artificial Intelligence (XAI) with churn prediction models, specifically using SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations). We apply these methods to machine learning models that use digital banking customer data, evaluating both how well they predict churn and how easy they are to understand for users and compliance teams. The study presents a framework to assess interpretability based on fidelity, stability, usability for stakeholders, and fairness. Our findings offer real insights into the balance between model accuracy and transparency, providing practical guidance for responsible use of AI in managing customer experiences. The study aims to promote ethical AI in finance by matching technical solutions with regulatory requirements and the need for human-centered understanding.

Educational Data Mining (EDM) supports early detection of learning difficulties by predicting student performance. However, machine learning models often operate as black boxes. Explainable Artificial Intelligence (XAI) helps to explain why black-box models produce specific predictions. This paper systematically reviews the past five years of research on XAI applications for interpreting student performance prediction in Computer Science and STEM education. We found that behavioral and academic performance data were the most commonly used features, with the main prediction goals focused on course failure risk or grades. This study also examined the application areas of XAI, revealing that the most common uses were global feature importance analysis, individual prediction explanations, and supporting interventions and decision-making. Moreover, we found that SHapley Additive exPlanations (SHAP) were the most frequently utilized XAI technique, predominantly applied at the global level, with limited use at the individual level. Furthermore, a research gap was identified in utilizing XAI to support course improvements, customize visualizations, and generate personalized recommendations. Addressing this gap could enable educators to provide personalized, data-driven guidance to better support individual students.

An algorithmic culture has emerged in which decisions affecting our daily life increasingly depend on automated systems such as machine learning. Developers of those systems strive for more accuracy, while at the same time demands for accountability increase. In many real-life applications, black box systems operate across domains like finance, health care, transportation, and criminal justice, often leaning towards more complex and less transparent machine learning (ML) models. Stakeholders responsible for an automating decision that can dramatically influence citizens' lives and regulate the operations of automated systems require information on why values are assigned and details of how the modelling decisions were made to build trust. Financial institutions are at the forefront of this trend and are slowly but creatively adopting these technologies to perform fundamentals such as financial audits, risk assessment, fraud detection, and customer scoring (

Introduction
Artificial Intelligence (AI) is only as powerful as the data that fuels it, and this book is your
comprehensive guide to understanding the critical data infrastructure that makes AI work.
AI has become a transformative force across industries, from healthcare and finance to retail and
manufacturing. However, while much attention is given to AI models and algorithms, the data that
feeds these systems is often overlooked. This book shifts the focus to the foundational elements of
AI—data architecture, storage, processing, and governance—so that organizations can effectively
harness the potential of AI. Even the most advanced AI models cannot deliver reliable results
without high-quality, well-structured data.
Part I lays the groundwork for understanding how AI has developed alongside advances in data
technology.
• Chapter 1: Introduction to Data for AI: This chapter introduces the book’s central
theme: the importance of data in AI. It highlights the three major pillars driving AI
adoption—computing power, data technology, and novel applications—while
emphasizing that data remains the most overlooked yet essential component.
• Chapter 2: Data Mining for AI: Explores the origins of data mining and its foundational
role in AI, detailing key methodologies like the CRISP-DM process and how data
preprocessing, cataloging, and visualizing support AI.
• Chapter 3: Data Challenges in Machine Learning: Addresses the technical debt
associated with ML systems, outlining common data dependencies, feedback loops, and
strategies to overcome these issues in AI development.
• Chapter 4: Deep Learning and Data Infrastructure: Examines the rise of deep learning,
the impact of Apache Spark, and the shift to data lakes that enabled more advanced AI
models, including Convolutional Neural Networks (CNNs).
• Chapter 5: ChatGPT and Large Language Models: Discusses the evolution of large
language models, the massive data requirements needed for training, and the challenges of
fine-tuning and deployment.
• Chapter 6: Data in Generative AI: Covers the specific data challenges of generative AI,
such as storage, movement, and ethical considerations related to training on vast datasets.
Part II provides practical guidance on managing and optimizing data for AI-driven organizations.
• Chapter 7: Modern Data Storage and Processing for AI: Reviews current trends in data
storage, including cloud-based solutions, edge computing, and data lake architectures
tailored for AI applications.
• Chapter 8: MDM and Data Quality for AI: Explores the importance of master data
management, the “garbage in, garbage out“ principle, and strategies for ensuring high data
quality in AI pipelines.
• Chapter 9: Ethical Data Management and Governance for AI: Discusses the critical role
of governance frameworks, compliance requirements, and the technology needed to
enforce ethical AI practices.
• Chapter 10: How Data Moves in AI-Powered Organizations: Provides insights into data
pipelines, orchestration, and real-time processing in AI-driven enterprises, ensuring data
is effectively utilized across systems.
• Chapter 11: Making AI Operational: Examines the deployment of AI systems, from
model integration to real-world application, with an emphasis on maintaining
performance over time.
• Chapter 12: Avoiding Common Pitfalls and The Future of AI: Highlights frequent
mistakes in AI projects and explores emerging trends in AI data infrastructure, preparing
organizations for the next wave of technological advancement.
This book offers a structured and practical approach to understanding data for AI by bridging
theoretical concepts with real-world applications. It equips readers with the insights and tools
necessary to build robust AI-driven systems that are efficient, ethical, and scalable.

Over 64 million people worldwide are affected by heart failure (HF), a condition that significantly raises mortality and medical expenses. In this study, we explore the potential of retinal optical coherence tomography (OCT) features as noninvasive biomarkers for the classification of heart failure subtypes: left ventricular heart failure (LVHF), congestive heart failure (CHF), and unspecified heart failure (UHF). By analyzing retinal measurements from the left eye, right eye, and both eyes, we aim to investigate the relationship between ocular indicators and heart failure using machine learning (ML) techniques. We conducted nine classification experiments to compare normal individuals against LVHF, CHF, and UHF patients, using retinal OCT features from each eye separately and in combination. Our analysis revealed that retinal thickness metrics, particularly ISOS-RPE and macular thickness in various regions, were significantly reduced in heart failure patients. Logistic regression, CatBoost, and XGBoost models demonstrated robust performance, with notable accuracy and area under the curve (AUC) scores, especially in classifying CHF and UHF. Feature importance analysis highlighted key retinal parameters, such as inner segment-outer segment to retinal pigment epithelium (ISOS-RPE) and inner nuclear layer to the external limiting membrane (INL-ELM) thickness, as crucial indicators for heart failure detection. The integration of explainable artificial intelligence further enhanced model interpretability, shedding light on the biological mechanisms linking retinal changes to heart failure pathology. Our findings suggest that retinal OCT features, particularly when derived from both eyes, have significant potential as non-invasive tools for early detection and classification of heart failure. These insights may aid in developing wearable, portable diagnostic systems, providing scalable solutions for personalized healthcare, and improving clinical outcomes for heart failure patients.

Chronic Kidney Disease (CKD) is a progressive condition that requires accurate diagnosis and staging for effective clinical management. Conventional CKD diagnosis relies on estimated Glomerular Filtration Rate (eGFR), a measure of kidney function derived from serum biomarkers such as serum creatinine (SCr) and cystatin C (SCysC). However, eGFR calculations may be inaccurate when applied to diverse patient populations. This study proposes a machine learning (ML) system that integrates regression-based eGFR estimation, metaheuristic optimization using the Grey Wolf Optimizer (GWO), and multi-class classification with various ML models to enhance CKD staging and classification. The model estimates eGFR using three established CKD Epidemiology Collaboration (CKD-EPI) equations incorporating SCr, SCysC, and their combined values. Regression models assess predictive performance, specifically Linear Regression (LR) and Support Vector Regression (SVR). SVR demonstrates superior performance compared to LR for CKD-EPI SCr-SCysC achieved a root mean squared error (RMSE) of 3.03, a mean absolute percentage error (MAPE) of 2.97%, and a coefficient of determination (R 2) score of 0.97. The application of GWO for hyperparameter tuning has resulted in a 37.3% reduction in root mean square error (RMSE), a 37.4% drop in mean absolute percentage error (MAPE), and a 2.06% improvement in R 2 to improve the precision of prediction. Once the model fine-tunes the eGFR estimations, it feeds them into various algorithms for CKD stage classification, including Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). Among these, XGBoost achieves the highest classification accuracy of 97.76%, along with an F1-score of 97.45%, demonstrating its effectiveness in CKD staging. Shapley Additive Explanations (SHAP) provide global and local feature importance insights, enhancing clinical decision-making and model transparency. Future research will validate the model using more extensive and more diverse datasets. Additionally, it will incorporate extra clinical parameters, including biomarkers and genetic data, to enhance the precision of CKD risk prediction. This research enhances AI-driven nephrology by providing a scalable, interpretable, and highly accurate solution for diagnosing and managing CKD. INDEX TERMS Chronic kidney diseases, CKD-EPI equation, cystatin C, glomerular filtration rate, serum creatinine, machine learning, explainable AI.

Alpay Algebra is introduced as a self-contained axiomatic framework with the ambition of serving as a universal foundation for mathematics. Developed in the spirit of Bourbaki's structural paradigm and Mac Lane's emphasis on form and function, Alpay Algebra posits a single abstract system from which diverse mathematical domains emerge. We present the precise axioms defining Alpay Algebra and develop its core constructs-including a recursive transformation operator ϕ, its transfinite iteration ϕ ∞ , an iterative state hierarchy χ λ , a limit object Ξ ∞ , and an evaluation functional ψ λ. From these primitives, we rigorously rebuild key fields: we derive fixed-point theorems and internal stability results, realize category theory within the algebra by interpreting compositional morphisms as iterative state transitions, recast homological algebra through cycles and invariants of the recursion, and outline an internal logic akin to topos theory grounded in stable truth values emerging from ψ λ. All definitions, theorems, and proofs are given entirely within the Alpay Algebra system without appeal to external frameworks. The development showcases how Alpay Algebra can subsume algebraic geometry, category theory, homological algebra, logic (including topos theory), and general structural mathematics under one unifying language. We conclude by highlighting new conjectures and problems that naturally arise from this universal algebraic perspective, underscoring the foundational depth and future potential of Alpay Algebra.

In this research paper, I argue that, with the new paradigm of deep learning, we should further incorporate evaluation, accountability and acceptance within the interface human-agent. As we shall see, this leads us to move from explainability to explanation.

This study aims to explore new therapeutic opportunities for histone deacetylase (HDAC) inhibitors by leveraging drug repurposing approaches and analyzing their bioactivity and molecular fingerprints. The methodology includes investigating drug repurposing opportunities for HDAC inhibitors, evaluating the bioactivity of repurposing drugs on HDAC enzymes, investigating the role of HDAC genes in therapeutic effects, and analyzing molecular fingerprints with explainable artificial intelligence (XAI) to identify structurally similar compounds with potential HDAC inhibitory activity. In this context, chemical compounds with IC50 (7903 compounds) and Inhibition (1084 compounds) standard types of HDAC genes reported to be associated with childhood acute leukemia were represented by molecular fingerprints. Regression and classification models were applied to the molecular fingerprints, and the results obtained were supported by XAI. All the study results were shared interactively on the website address arayuz-r89zld.streamlit.app/ by designing a simulation environment. The influence of molecular fingerprints on the models and their impact on potential drug development in childhood acute leukemia were evaluated using XAI techniques, particularly through the analysis of SHAP values. The study contributes to the literature on the use of XAI technology in drug repurposing studies, especially in cancer, the study of molecular properties, and the active use of XAI in drug repurposing studies.

In order to get a correct diagnosis and choose the best treatment options before it becomes deadly, early detection and classification of pancreatic tumours are essential. Grading can be a tedious and time-consuming process for experts and doctors when the case is complex. In such cases, experts usually look at the tumour and pinpoint its exact position. Moreover, it could be required to compare the tumor's cells to those in the vicinity. The end goal is to confirm that the growth is a tumour and, if possible, to ascertain the exact type and grade of the tumour. However, due to the high amounts of weights sent and received from the client-side trained models, federated learning techniques incur substantial communication overhead. This study aims to address this problem by introducing a unified framework that integrates the inherent capabilities of Federated Learning (FL) with the unique characteristics of the Grey Wolf Optimisation algorithm. The pancreatic tumour dataset is used to evaluate the GWO-enabled FL framework. The proposed model was more network efficient, performed better in data imbalance scenarios, and led to lower communication costs than the currently available federated average model. Following validation, the proposed framework attained a prediction accuracy of 98.9 %. For pancreatic tumour classification, the data obtained from the proposed system can be a useful component.

Fruit research now has reached a new dimension thanks to machine learning, which produces actionable insights for further exploration by practitioners in the agricultural domain. In order to automatically categorize the edibility of date fruit, we investigated various types of date fruits and used explainable artificial intelligence (XAI) techniques combined with machine learning-based methods to effectively classify and explain the classification task. Our result shows that with a formidable accuracy ranging within the 90-92 percentile for seven methods, including boosting, bagging, support vector machine (SVM), k-nearest neighbor (KNN), and MLPs, the machine learning methods combined with Local Interpretable Model-Agnostic Explanations (LIME) based XAI provides better actionable insights which can be utilized by domain experts and stakeholders to produce and supply quality fruits particularly date fruits thus contributing in broader perspectives with respect to this dynamically evolving domain. The implementation of the investigation and experiment is available in Github. 1

This Research Full Paper focuses on predicting learning performance using machine learning algorithms and interpreting the results using Explainable Machine Learning (EML) techniques. The study compared a comprehensive set of machine learning algorithms, including Logistic Regression, Decision Trees, AdaBoost, XGBoost, SVM, and KNN. The performance of these algorithms in predicting students' final grades in a course was accessed using various evaluation metrics. Our study used feature selection to identify the most relevant predictors to enhance predictive accuracy, implemented the Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance, and performed hyperparameter optimization to find the most effective model settings. This comprehensive approach improved the predictive accuracy of our models over previous studies. Additionally, the importance of early prediction in identifying at-risk students was explored, with models demonstrating promising accuracy at the first checkpoint of the course. Departing from traditional machine learning research that often focused on model performance, our study integrated the EML technique of Shapley Additive exPlanations (SHAP), which is grounded on the theoretical framework of Game Theory, to facilitate the interpretation of the predictive outcomes. This approach offered an explanatory perspective on the key factors influencing model decisions. By contributing to the predictability and interpretability of student performance, this research enriched the field of Educational Data Mining (EDM) and enhanced the understanding of student learning trajectories.

This study explores vulnerability management and risk assessment through data analysis and machine learning techniques. In the first phase, a preliminary analysis was conducted to understand the distribution of vulnerabilities based on key metrics such as Attack Vector, Impact Score, Base Score, and Exploitability Score. Statistical methods, correlation analysis, and clustering techniques were applied to uncover relationships among these attributes. Results indicated strong positive correlations between Impact Score and Base Score, as well as among the sub-metrics of Impact Score (Confidentiality, Integrity, and Availability).

In the second phase, supervised machine learning models were employed to enhance risk assessment and vulnerability classification. A Decision Tree model and Multinomial Logistic Regression were implemented to classify vulnerabilities based on attributes such as Attack Vector, Attack Complexity, and Base Severity. The Decision Tree model demonstrated superior performance in predicting vulnerability prioritization, with high precision and recall scores. Findings suggest that combining correlation analysis with supervised machine learning can significantly improve the vulnerability prioritization process, aiding cybersecurity analysts' inefficient decision-making.

Purpose: The article illustrates that data integration will play a pivotal role in value-based healthcare, amplifying the power of the best cost management and improving outcomes. It brings into focus the fundamental systems, methodologies, and best practices for integrating diverse sources of data that enable insight into action, driving cost efficiency and quality care. Methodology: The Article systematically reviews foundational systems and technologies essential for cost-of-care analytics in healthcare, with a focus on data integration frameworks such as HL7 and FHIR [9]. Different strategies for integrating these disparate data sources have been explored, including data warehousing and cloud-based solutions. Major challenges identified and analyzed in this article include data privacy, standardization, and organizational resistance to implementation. On the other hand, the best practices for the implementation process have been explored, including staged adoption, stakeholder engagement, and investment in leading analytics tools. Findings: Data integration in value-based healthcare has to be undertaken for both cost efficiency and improved patient outcomes. This integration process does need interoperable systems and centralized repositories [14] [31]. Privacy concerns, standardization, and such challenges need to be overcome through phased implementation, stakeholder engagement, and advanced analytics tools. Unique contribution to theory, practice and policy: The article contributes to the theory by placing data integration at the center of value-based health care, thereby connecting analytics to cost and outcome optimization. In practice, it would provide actionable methodologies-including interoperable frameworks for data standardization and phased implementation-to ensure seamless integration. It illustrates best practices that outline how to overcome operational hurdles, such as resistance to change and fragmented systems. From a policy perspective, emphasis is placed on the adherence to regulations such as HIPAA, ensuring that patient data are used securely and in an ethical manner. Putting these findings together will bridge the gaps in theory, practice, and policy for cost-effective quality care.

The XAI concept was launched by the DARPA in 2016 in the context of model learning from data with deep learning methods. Although the machine learning community quickly took up on the topic, other communities have also included explanation in their research agenda (e.g. Case Based Reasoning, Planning, Decision Support, Emerging Systems, Robotics, Internet of Things). The question of explanation, which is at the center of philosophical research works, has been revisited during the last decades. The humanities community insists on the fact that explanation is above all a process that develops in the context of the search for explanation and cannot be completely defined a priori. In this contribution, we propose 1) to broaden the question of explanation to any type of situation in which users exploit the possibilities of decision support agents for their own decisions, in the context of their task, and within the framework of their activities and responsibilities and 2) to consider an instrumentation of digital devices, able to manage dynamic explanation agents associated to corresponding decision support agents. We denote this evolution "UXAI" (User eXplainable Artificial Intelligence) because we consider that users should be the main actors in the dynamics of any explanation process.

The XAI concept was launched by the DARPA in 2016 in the context of model learning from data with deep learning methods. Although the machine learning community quickly took up on the topic, other communities have also included explanation in their research agenda (e.g. Case Based Reasoning, Planning, Decision Support, Emerging Systems, Robotics, Internet of Things). The question of explanation, which is at the center of philosophical research works, has been revisited during the last decades. The humanities community insists on the fact that explanation is above all a process that develops in the context of the search for explanation and cannot be completely defined a priori. In this contribution, we propose 1) to broaden the question of explanation to any type of situation in which users exploit the possibilities of decision support agents for their own decisions, in the context of their task, and within the framework of their activities and responsibilities and 2) to consider an ...

Artificial Intelligence (AI) is adopted in many businesses. However, adoption lacks behind for use cases with regulatory or compliance requirements, as validation and auditing of AI is still unresolved. AI's opaqueness (i.e., "black box") makes the validation challenging for auditors. Explainable AI (XAI) is the proposed technical countermeasure that can support validation and auditing of AI. We developed an XAI based validation approach for AI in sensitive use cases that facilitates the understanding of the system's behaviour. We conducted a case study in pharmaceutical manufacturing where strict regulatory requirements are present. The validation approach and an XAI prototype were developed through multiple workshops and then tested and evaluated with interviews. Our approach proved suitable to collect the required evidence for a software validation, but requires additional efforts compared to a traditional software validation. AI validation is an iterative process and clear regulations and guidelines are needed.

Chronic kidney disease (CKD) is a major worldwide health problem, affecting a large proportion of the world’s population and leading to higher morbidity and death rates. The early stages of CKD sometimes present without visible symptoms, causing patients to be unaware. Early detection and treatments are critical in reducing complications and improving the overall quality of life for people afflicted. In this work, we investigate the use of an explainable artificial intelligence (XAI)-based strategy, leveraging clinical characteristics, to predict CKD. This study collected clinical data from 491 patients, comprising 56 with CKD and 435 without CKD, encompassing clinical, laboratory, and demographic variables. To develop the predictive model, five machine learning (ML) methods, namely logistic regression (LR), random forest (RF), decision tree (DT), Naïve Bayes (NB), and extreme gradient boosting (XGBoost), were employed. The optimal model was selected based on accuracy and area under the curve (AUC). Additionally, the SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) algorithms were utilized to demonstrate the influence of the features on the optimal model. Among the five models developed, the XGBoost model achieved the best performance with an AUC of 0.9689 and an accuracy of 93.29%. The analysis of feature importance revealed that creatinine, glycosylated hemoglobin type A1C (HgbA1C), and age were the three most influential features in the XGBoost model. The SHAP force analysis further illustrated the model’s visualization of individualized CKD predictions. For further insights into individual predictions, we also utilized the LIME algorithm. This study presents an interpretable ML-based approach for the early prediction of CKD. The SHAP and LIME methods enhance the interpretability of ML models and help clinicians better understand the rationale behind the predicted outcomes more effectively.

This study aims to explore new therapeutic opportunities for histone deacetylase (HDAC) inhibitors by leveraging drug repurposing approaches and analyzing their bioactivity and molecular fingerprints. The methodology includes investigating drug repurposing opportunities for HDAC inhibitors, evaluating the bioactivity of repurposing drugs on HDAC enzymes, investigating the role of HDAC genes in therapeutic effects, and analyzing molecular fingerprints with explainable artificial intelligence (XAI) to identify structurally similar compounds with potential HDAC inhibitory activity. In this context, chemical compounds with IC50 (7903 compounds) and Inhibition (1084 compounds) standard types of HDAC genes reported to be associated with childhood acute leukemia were represented by molecular fingerprints. Regression and classification models were applied to the molecular fingerprints, and the results obtained were supported by XAI. All the study results were shared interactively on the website address arayuz-r89zld.streamlit.app/ by designing a simulation environment. The influence of molecular fingerprints on the models and their impact on potential drug development in childhood acute leukemia were evaluated using XAI techniques, particularly through the analysis of SHAP values. The study contributes to the literature on the use of XAI technology in drug repurposing studies, especially in cancer, the study of molecular properties, and the active use of XAI in drug repurposing studies.

Explainable artificial intelligence (XAI) uses artificial intelligence (AI) tools
and techniques to build interpretability in black-box algorithms. XAI
methods are classified based on their purpose (pre-model, in-model, and
post-model), scope (local or global), and usability (model-agnostic and
model-specific). XAI methods and techniques were summarized in this
paper with real-life examples of XAI applications. Local interpretable
model-agnostic explanations (LIME) and shapley additive explanations
(SHAP) methods were applied to the moral dataset to compare the
performance outcomes of these two methods. Through this study, it was
found that XAI algorithms can be custom-built for enhanced model-specific
explanations. There are several limitations to using only one method of XAI
and a combination of techniques gives complete insight for all stakeholders.

Explainable Artificial Intelligence (XAI) is a rapidly evolving field aimed at making AI systems more interpretable and transparent to human users. As AI technologies become increasingly integrated into critical sectors such as healthcare, finance, and autonomous systems, the need for explanations behind AI decisions has grown significantly. This survey provides a comprehensive review of XAI techniques, categorizing them into post-hoc and intrinsic methods, and examines their application in various domains. Additionally, the paper explores the major challenges in achieving explainability, including balancing accuracy with interpretability, scalability, and the trade-off between transparency and complexity. The survey concludes with a discussion on the future directions of XAI, emphasizing the importance of interdisciplinary approaches to developing robust and interpretable AI systems.

Patients with chronic kidney disease (CKD) face a high risk of cardiovascular death, yet accurately predicting this risk remains challenging. This study aims to develop an interpretable machine learning (ML) model to predict 10-year cardiovascular mortality in CKD patients using SHAP explainers. [1]Six ML models were created and tested, with the best model selected for prediction and patient categorization. Survival rates were analyzed using log-rank tests on Kaplan-Meier curves, and Cox regression was employed to explore the relationship between ML-predicted risk scores and mortality. The chosen autoencoder (AE) model demonstrated superior performance, with higher ML scores[2] significantly correlating with increased cardiovascular mortality risk. Key determinants such as age, high blood pressure, C-reactive protein, and serum creatinine were identified. The ML-driven tool showcased high accuracy in determining the 10-year cardiovascular mortality risk for CKD patients, offering valuable insights for individual risk assessments.

The communication between robots/agents and humans is a challenge, since humans are typically not capable of understanding the agent's state of mind. To overcome this challenge, this paper relies on recent advances in the domain of eXplainable Artificial Intelligence (XAI) to trace the decisions of the agents, increase the human's understandability of the agents' behavior, and hence improve efficiency and user satisfaction. In particular, we propose a Human-Agent EXplainability Architecture (HAEXA) to model human-agent explainability. HAEXA filters the explanations provided by the agents to the human user to reduce the user's cognitive load. To evaluate HAEXA, a human-computer interaction experiment is conducted, where participants watch an agent-based simulation of aerial package delivery and fill in a questionnaire that collects their responses. The questionnaire is built according to XAI metrics as established in the literature. The significance of the results is verified using Mann-Whitney U tests. The results show that the explanations increase the understandability of the simulation by human users. However, too many details in the explanations overwhelm them; hence, in many scenarios, it is preferable to filter the explanations.

Motivated by the apparent societal need to design complex autonomous systems whose decisions and actions are humanly intelligible, the study of explainable artificial intelligence, and with it, research on explainable autonomous agents has gained increased attention from the research community. One important objective of research on explainable agents is the evaluation of explanation approaches in humancomputer interaction studies. In this demonstration paper, we present a way to facilitate such studies by implementing explainable agents and multi-agent systems that i) can be deployed as static files, not requiring the execution of server-side code, which minimizes administration and operation overhead, and ii) can be embedded into web front ends and other JavaScript-enabled user interfaces, hence increasing the ability to reach a broad range of users. We then demonstrate the approach with the help of an application that was designed to assess the effect of different explainability approaches on the human intelligibility of an unmanned aerial vehicle simulation.

Iron (Fe) chelating medicines and Histone deacetylase (HDAC) inhibitors are two therapy options for hereditary Friedreich's Ataxia that have been shown to improve clinical results (FA). Fe chelation molecules can minimize the quantity of stored Fe, and HDAC inhibitors can boost the expression of the Frataxin (FXN) gene in enhancing FA. A complete quantitative structure-activity relationship (QSAR) search of inhibitors from the ChEMBL database is reported in this paper, which includes 437 compounds for Fe chelation and 1,354 compounds for HDAC inhibitors. For further investigation, the IC50 was chosen as the unit of bioactivity, and following data refinement, a final dataset of 436 and 1,163 compounds for Fe chelation and HDAC inhibition, respectively, was produced. The Random Forest (RF) technique was used to generate models (train R2 score, 0.701 and 0.892; test R2 score 0.572 and 0.460, for Fe and HDAC, respectively). The models created using the PubChem fingerprint were the strongest of the 12 fingerprint kinds; hence that feature was chosen for interpretation. The results showed the importance of properties related to nitrogen-containing functional groups (SHAP value of PubchemFP656 is -0.29) and aromatic rings (SHAP value of PubchemFP12 is -0.16). As a result, we explained the effect of the molecular fingerprints on the models and the impact on possible drugs that can be developed for FA with artificial intelligence (XAI), which can be explained through SHAP (Shapley Additive Explanations) values. Model scripts and fingerprinting methods are also available at https://github.com/tissueandcells/XAI.

Artificial Intelligence has now taken a full-fledged role in healthcare and has started driving innovations not only in diagnostics and treatment planning but also in patient monitoring and operational efficiency. This will enable complex medical data analysis, extracting patterns and insights that no human is capable of. However, most of these models are per se opaque-that is, the so-called "black-box" problem-there are still great challenges in areas such as transparency, trust, and ethical applications in a clinical setting. This lack of interpretability can stand in the way of acceptance or integration for AI technologies when issues of understanding and accountability are relevant. Explainable AI solves these problems by making real artificial intelligence decisions understand the decisions made to humans. XAI techniques offer well-understandable and interpretable explanations of the models with minimum degradations in performance. This review article explains in detail the critical role of XAI in healthcare, underpinning how this field can bring more transparency into AI applications. We explain some of the current methods of XAI: modelagnostic techniques like LIME and SHAP, interpretable models relating to decision trees and linear models, and visualization techniques like saliency maps and mechanisms of attention.

Today, with the development of artificial intelligence, its application in different areas, including production and operations, has expanded. Explainable artificial intelligence (XAI) is a new research topic that has emerged with the development of artificial intelligence. This study aimed to investigate the applications of XAI in production and operations using the systematic review approach. For this purpose, a systematic review of the most recent studies published in the Science Direct, Scopus, and Emerald knowledge bases was conducted. After screening through different stages, 29 articles were reviewed and analyzed. The results showed that publications on XAI have been on an upward trend in recent years, with a significant increase observed from 2021 to 2024. Also, the fields of engineering, production, decision-making, and computer science are the major areas in which recent studies have been published. The results also suggested that the largest scope of XAI application was observed at the organizational level, followed by the industrial level. Based on the findings, the fields of production and operations, followed by logistics and supply chain, were the most frequently studied areas. Regarding the methods used, the SHAP method was the most commonly applied method in the XAI studies, followed by Integrated Gradient and SVM methods. In general, the results of this study showed that XAI is a new field of research that is gradually developing in terms of methodology and areas of application.

The field of explainable artificial intelligence (XAI) aims to explain the decisions of DNNs. Complete DNN explanations accurately reflect the inner workings of the DNN while interpretable explanations are easy for humans to understand. Developing methods for explaining the representations learned by DNNs that are both complete and interpretable is a grand challenge in the field of XAI. This thesis makes contributions to the field of XAI by proposing and evaluating novel methods for studying DNN representations. During forward propagation, each DNN layer non-linearly transforms the input space in some way that is useful for minimizing a loss function. To understand how DNNs represent their inputs, this work develops methods to examine each DNN layer’s activation space. The first article contributes an unsupervised framework for identifying and interpreting “tuning dimensions” in the activation space of DNN layers. The method consists of fitting a dimensionality reduction model to a ...

Background: This study utilized advanced Artificial Intelligence (AI) techniques to develop predictive models for legume crop yields
in the context of climate change scenarios. With the escalating challenges posed by climate change, accurately forecasting agricultural
outcomes is imperative for sustainable food production.

Methods: Utilizing an extensive dataset comprising legume crop yields, climate change forecasts and relevant environmental factors,
this study employs advanced machine learning techniques such as XGBoost to create strong predictive models. The analysis
encompasses diverse climate change scenarios to assess the resilience of legume crops under varying environmental conditions.

Result: Results indicate a significant enhancement in predictive accuracy compared to conventional models, demonstrating the
efficacy of AI in anticipating legume crop yields amidst climatic uncertainties. The presented work not only improves the precision of
agricultural predictive modeling but also underscores the vital role of AI in mitigating the detrimental effects of climate change on
food security. The agriculture industry faces changing weather patterns, thus using AI-powered prediction models becomes essential
for making well-informed decisions and implementing sustainable farming methods.

Artificial Intelligence is making significant inroads into various aspects of business and life bringing the transformation in many ways. The convolution of technology in finance is often called FINTECH rapidly growing area of transformation. In the FINTECH industry, AI can automate several financial processes and services such as fraud detection, customer services, credit assessment, price predictions, customer churning, trading services, risk management, underwriting, market forecasting. These processes and services are critical to financial sectors such as banking, insurance, currency, stock and commodity markets, wealth management, payment clearing houses, payment regulators etc. Regulations control these processes and should be transparent in their operations. AI models are inherently opaque in their outcomes and unable to be fully plugged into the financial processes and services. Explainable AI is the key area of research that can help to provide transparency to enable these AI models as fully operational business models to automate financial products and services. In this paper we will broadly outline the framework of explainable artificial intelligence (XAI) in finance sectors and services. We then look into one use case of credit assessment and develop an XAI framework to provide transparent outcomes from the AI models.

Explainable AI (XAI) has emerged as a critical field in artificial intelligence, addressing the "black box" nature of complex machine learning models. This article explores the importance of transparency in AI decision-making, the techniques used to achieve explainability, and the implications for various sectors. We examine the current state of XAI, its applications in healthcare, finance, and other critical areas, and discuss the ethical and regulatory considerations surrounding transparent AI. The paper concludes with an analysis of XAI's contribution to the broader field of AI and its potential future developments.

The rapid advancement of artificial intelligence (AI) has led to its widespread adoption across various domains. One of the most important challenges faced by AI adoption is to justify the outcome of the AI model. In response, explainable AI (XAI) has emerged as a critical area of research, aiming to enhance transparency and interpretability in AI systems. However, existing XAI methods facing several challenges, such as complexity, difficulty in interpretation, limited applicability, and lack of transparency. In this paper, we discuss current challenges using SHAP and LIME metrics being popular methods for explainable AI and then present a novel approach for developing an explainable AI framework that addresses these limitations. This novel approach uses simple techniques and understandable human explanations to provide users with clear and interpretable insights into AI model behavior. Key components of this approach include model-agnostic interpretability, a newly developed explainable factor overcoming the challenges of current XAI methods and enabling users to understand the decision-making process of AI models. We demonstrate the effectiveness of the new approach through a case study and evaluate the framework's performance in terms of interpretability. Overall, the new approach offers enhanced transparency and trustworthiness in AI systems across diverse applications.

Artificial Intelligence is making significant inroads into various aspects of business and life bringing the transformation in many ways. The convolution of technology in finance is often called FINTECH rapidly growing area of transformation. In the FINTECH industry, AI can automate several financial processes and services such as fraud detection, customer services, credit assessment, price predictions, customer churning, trading services, risk management, underwriting, market forecasting. These processes and services are critical to financial sectors such as banking, insurance, currency, stock and commodity markets, wealth management, payment clearing houses, payment regulators etc. Regulations control these processes and should be transparent in their operations. AI models are inherently opaque in their outcomes and unable to be fully plugged into the financial processes and services. Explainable AI is the key area of research that can help to provide transparency to enable these AI models as fully operational business models to automate financial products and services. In this paper we will broadly outline the framework of explainable artificial intelligence (XAI) in finance sectors and services. We then look into one use case of credit assessment and develop an XAI framework to provide transparent outcomes from the AI models.

One of the most critical aspects of a software piece is its vulnerabilities. Regardless of the years of experience, type of project, or the size of the team, it is impossible to avoid introducing vulnerabilities while developing or maintaining software. This aspect becomes crucial when the software is deployed in production or released to the final users. At that point finding vulnerabilities becomes a race between the developers and malicious intruders, whoever finds it first can either exploit it or fix it. Acknowledging this situation and using the tools and standards that we have available in the field, such as common vulnerability exposures and common vulnerability scoring systems, and based on modern researches, in this study, we propose to have an approach different from the common practices of manual classification, using a 2-layer convolutional neuronal network (CNN) to automatize the classification of vulnerabilities, speeding up this process and enabling developers to have a faster response towards vulnerabilities, producing safer software. The experimental results obtained in this study suggest that pre-trained word embeddings contributed to an increase in accuracy of approximately 2% and the overall accuracy become 0.816%.

Edge devices that operate in real-world environments are subjected to unpredictable conditions caused by environmental forces such as wind and uneven surfaces. Since most edge systems such as autonomous vehicles exhibit dynamic properties, it is clear that reinforcement learning can be a powerful tool for improving system accuracy. Successful maintenance of the position of a vehicle in such environments has been recently achieved with the aid of Deep Reinforcement Learning (DRL) that dynamically adjusts the Reconfigurable Wireless Network (RWN) response. Deep Neural Networks (DNNs) are often seen as black boxes, as neither the acquired knowledge nor the decision rationale can be explained. This raises issues with the transparency and trustworthiness of the system because the underlying AI models are not governed by any mathematical or physics laws. In this paper, we explain the process of a DNN utilisation in an autonomous dynamic positioning system by gauging reactions of the DNN to predefined constraints and capturing the associated conditions that influence the DNN in a time series. We introduce a novel digitisation technique that reduces interesting patterns of time series data into single-digits to obtain a cross comparable view of the conditions. By analysing the clusters formed on this cross comparable view, we discovered multiple intensities of environmental conditions spanning across 44% of moderate conditions and 33% and 23% of harsh and mild conditions respectively. Our analysis showed that the proposed system can provide stable responses to uncertain conditions by predicting randomness.

An Advisory System using Sentiment Analysis (ASSA) is proposed to analyze student reviews and assist faculty and university officials in addressing challenging areas of teaching and learning. In addition to analyzing student comments to determine sentiment polarity, the emotions communicated, and contentment versus discontent, the proposed system also enables the advisor to provide their comments in response to the student's feedback. ASSA will generate recommendations based on input from both the students and their advisors. ASSA for improved student feedback and academic advising has a high accuracy rate that supports and aids academic advisors in their attempts to provide reliable, accurate, and continuous advising services to their students. Also, this system aims to execute the classification of review data and perform prescriptive analysis with promising outcomes. ASSA has proven to be an effective tool for academic advisors, enabling them to analyze student feedback, offer appropriate recommendations, and enhance the overall advising experience.

In recent years, deep learning has become prevalent to solve applications from multiple domains. Convolutional Neural Networks (CNNs) particularly have demonstrated state of the art performance for the task of image classification. However, the decisions made by these networks are not transparent and cannot be directly interpreted by a human. Several approaches have been proposed to explain to understand the reasoning behind a prediction made by a network. In this paper, we propose a topology of grouping these methods based on their assumptions and implementations. We focus primarily on white box methods that leverage the information of the internal architecture of a network to explain its decision. Given the task of image classification and a trained CNN, this work aims to provide a comprehensive and detailed overview of a set of methods that can be used to create explanation maps for a particular image, that assign an importance score to each pixel of the image based on its contribution to the decision of the network. We also propose a further classification of the white box methods based on their implementations to enable better comparisons and help researchers find methods best suited for different scenarios.

Recently, the eXplainable AI (XAI) research community has focused on developing methods making Machine Learning (ML) predictors more interpretable and explainable. Unfortunately, researchers are struggling to converge towards an unambiguous definition of notions such as interpretation, or, explanation-which are often (and mistakenly) used interchangeably. Furthermore, despite the sound metaphors that Multi-Agent System (MAS) could easily provide to address such a challenge, and agent-oriented perspective on the topic is still missing. Thus, this paper proposes an abstract and formal framework for XAI-based MAS, reconciling notions, and results from the literature.

With the recent successes of black-box models in Artificial Intelligence (AI) and the growing interactions between humans and AIs, explainability issues have risen. In this article, in the context of high-stake applications, we propose an approach for explainable classification and annotation of images. It is based on a transparent model, whose reasoning is accessible and human understandable, and on interpretable fuzzy relations that enable to express the vagueness of natural language. The knowledge about relations is set beforehand by an expert and thus training instances do not need to be annotated. The most relevant relations are extracted using a fuzzy frequent itemset mining algorithm in order to build rules, for classification, and constraints, for annotation. We also present two heuristics that make the process of evaluating relations faster. Since the strengths of our approach are the transparency of the model and the interpretability of the relations, an explanation in natural language can be generated. Supported by experimental results, we show that, given a segmentation of the input, our approach is able to successfully perform the target task and generate explanations that were judged as consistent and convincing by a set of participants.

This panel will discuss the problems of bias and fairness in organizational use of AI algorithms. The panel will first put forth key issues regarding biases that arise when AI algorithms are applied to organizational processes. We will then propose a sociotechnical approach to bias mitigation. We will further share proposals for companies and policymakers on improving AI algorithmic fairness within organizations and bias mitigation. The panel will bring together scholars examining social and technical aspects of bias and its mitigation, from the perspective of information systems, ethics, machine learning, robotics, and human capital. The panel will end with an open discussion of where the field of information systems can step in to guide fairness and ethical use in AI algorithms in the coming years.

AI tools are becoming more commonly used in a variety of application domains. In this paper, we describe a system named FATE that combines state of the art AI tools. The goal of the FATE system is decision support with use of ongoing human-AI co-learning, explainable AI and fair, bias-free and secure usage of data. These topics are societally very relevant for the update of AI-based support systems, but the manner in which to bring these together into a working system is far from trivial. We describe the various functional components that comprise the system, share our experience with the set-up of such a system, explain how it can be used in a variety of use cases, taking into account multiple user roles. Finally, we reflect on this and provide an outlook for the continuation of this development.

A governance framework for algorithmic accountability and transparency Algorithmic systems are increasingly being used as part of decision-making processes in both the public and private sectors, with potentially significant consequences for individuals, organisations and societies as a whole. Algorithmic systems in this context refer to the combination of algorithms, data and the interface process that together determine the outcomes that affect end users. Many types of decisions can be made faster and more efficiently using algorithms. A significant factor in the adoption of algorithmic systems for decision-making is their capacity to process large amounts of varied data sets (i.e. big data), which can be paired with machine learning methods in order to infer statistical models directly from the data. The same properties of scale, complexity and autonomous model inference however are linked to increasing concerns that many of these systems are opaque to the people affected by their use and lack clear explanations for the decisions they make. This lack of transparency risks undermining meaningful scrutiny and accountability, which is a significant concern when these systems are applied as part of decision-making processes that can have a considerable impact on people's human rights (e.g. critical safety decisions in autonomous vehicles; allocation of health and social service resources, etc.). This study develops policy options for the governance of algorithmic transparency and accountability, based on an analysis of the social, technical and regulatory challenges posed by algorithmic systems. Based on a review and analysis of existing proposals for governance of algorithmic systems, a set of four policy options are proposed, each of which addresses a different aspect of algorithmic transparency and accountability: 1. awareness raising: education, watchdogs and whistleblowers; 2. accountability in public-sector use of algorithmic decision-making; 3. regulatory oversight and legal liability; and 4. global coordination for algorithmic governance.