XAI - Explainable Artificial Intelligence

De contratos a sentenças judiciais, de diagnósticos médicos à criação artística, a inteligência artificial (IA) já participa ativamente da vida humana – muitas vezes por meio de algoritmos complexos e incompreensíveis. Mas como confiar em sistemas que não conseguem justificar as suas próprias escolhas? E como poderemos contestar decisões algorítmicas injustas ou preconceituosas? Nesta obra, Otávio Morato e Dierle Nunes apresentam a explicabilidade como princípio norteador para o uso mais ético e responsável da IA. Com linguagem acessível e dezenas de esquemas visuais, os autores articulam direito e tecnologia para mostrar que os algoritmos não precisam ser caixas-pretas. Trata-se de entender, questionar e democratizar as decisões automatizadas que, em certa medida, já nos governam.

This paper redefines product quality in educational technology, especially within digital health and e-learning systems, as more than uptime or usability. In high-stakes contexts, quality must be measured by a system’s ability to deliver safe, effective, and trustworthy learning experiences over time. Drawing on leadership experience at Lenus eHealth, the article outlines how engineered testing frameworks, automation-first strategies, and content validation pipelines transformed quality assurance from reactive bug-finding to a proactive, metrics-driven culture. Reported outcomes included reduced support tickets, increased team velocity, and higher user satisfaction. The discussion emphasizes the need for scenario-based testing, bias mitigation, explainability, and privacy as core design principles. Through real-world examples, the paper demonstrates how product quality, when embedded into development lifecycles and organizational leadership, becomes not just a safeguard but a strategic driver of innovation, compliance, and client trust in digital health and EdTech platforms.

Children learn continually by asking questions about the concepts they are most curious about. With robots becoming an integral part of our society, they must also learn unknown concepts continually by asking humans questions. This paper presents a novel framework for curiosity-driven online learning of objects. The paper utilizes a recent stateof-the-art approach for continual learning and adapts it for online learning of objects. The paper further develops a selfsupervised technique to find most of the uncertain objects in an environment by utilizing an internal representation of previously learned classes. We test our approach on a benchmark dataset for continual learning on robots. Our results show that our curiosity-driven online learning approach beats random sampling and softmax-based uncertainty sampling in terms of classification accuracy and the total number of classes learned.

During the early phase of the COVID-19 pandemic, reverse transcriptase-polymerase chain reaction (RT-PCR) testing faced limitations, prompting the exploration of machine learning (ML) alternatives for diagnosis and prognosis. Providing a comprehensive appraisal of such decision support systems and their use in COVID-19 management can aid the medical community in making informed decisions during the risk assessment of their patients, especially in low-resource settings. Therefore, the objective of this study was to systematically review the studies that predicted the diagnosis of COVID-19 or the severity of the disease using ML. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA), we conducted a literature search of MEDLINE (OVID), Scopus, EMBASE, and IEEE Xplore from January 1 to June 31, 2020. The outcomes were COVID-19 diagnosis or prognostic measures such as death, need for mechanical ventilation, admission, and acute respiratory distress syndrome. We included peer-reviewed observational studies, clinical trials, research letters, case series, and reports. We extracted data about the study's country, setting, sample size, data source, dataset, diagnostic or prognostic outcomes, prediction measures, type of ML model, and measures of diagnostic accuracy. Bias was assessed using the Prediction model Risk Of Bias ASsessment Tool (PROBAST). This study was registered in the International Prospective Register of Systematic Reviews (PROSPERO), with the number CRD42020197109. The final records included for data extraction were 66. Forty-three (64%) studies used secondary data. The majority of studies were from Chinese authors (30%). Most of the literature (79%) relied on chest imaging for prediction, while the remainder used various laboratory indicators, including hematological, biochemical, and immunological markers. Thirteen studies explored predicting COVID-19 severity, while the rest predicted diagnosis. Seventy percent of the articles used deep learning models, while 30% used traditional ML algorithms. Most studies reported high sensitivity, specificity, and accuracy for the ML models (exceeding 90%). The overall concern about the risk of bias was "unclear" in 56% of the studies. This was mainly due to concerns about selection bias. ML may help identify COVID-19 patients in the early phase of the pandemic, particularly in the context of chest imaging. Although these studies reflect that these ML models exhibit high accuracy, the novelty of these models and the biases in dataset selection make using them as a replacement for the clinicians' cognitive decision-making questionable. Continued research is needed to enhance the robustness and reliability of ML systems in COVID-19 diagnosis and prognosis.

This study examines how blockchain transparency and smart-contract automation, paired with anomaly-detection models, support early detection and calibrated deterrence of manipulation in cryptocurrency markets. While transparent ledgers and rule-based execution raise the likelihood that irregular activity is flagged and investigated, they do not prevent fraud; our emphasis is detection, deterrence, and post-incident support. We analyze a long-horizon Bitcoin panel using rolling z-score screens and Isolation Forest to surface anomalies consistent with manipulative trading. We fix a false-positive budget ex-ante and evaluate capacity-aware performance (precision@k, PR-AUC, lead time), archiving time-stamped evidence bundles for auditability. Alerts cluster around episodes consistent with pump-and-dump behavior, large-holder moves, and event-driven dislocations, improving investigative triage without prevention claims. The framework provides actionable guidance for exchanges and regulators seeking to strengthen market integrity through auditable records and model-based alerts, and we release a human-inthe-loop agentic AI application that automates ingestion, screening, ranking, and auditable export.

Artificial Intelligence (AI) has become a transformative force in social science research, enabling the analysis of large-scale, heterogeneous data to uncover latent patterns and predict complex human behaviors. Among AI's core methodologies, computer vision has evolved far beyond its early role in data acquisition to now encompass sophisticated systems capable of interpreting, analyzing, and synthesizing visual information across a range of socially relevant contexts. By integrating advanced image processing, machine learning, and computer graphics, computer vision empowers interdisciplinary investigations in psychology, sociology, and economics, modeling phenomena such as decision-making, emotional expression, and social interaction at unprecedented scales and resolutions. This paper presents a comprehensive survey of the state of the art in computer vision applications within the social sciences, with particular emphasis on recent breakthroughs in algorithms and enabling technologies that facilitate automated visual understanding. Key topics include object detection, facial recognition, scene understanding, and predictive modeling, which collectively underpin impactful applications in healthcare, autonomous systems, surveillance, and digital media. To structure this rapidly expanding domain, we propose a conceptual framework that organizes the field into four foundational pillars: image processing, object recognition, adaptive machine learning, and computer graphics. Each pillar contributes critical capabilities such as feature extraction, quality enhancement, semantic interpretation, and photorealistic rendering-functions that are increasingly pivotal in addressing contemporary social challenges. By critically evaluating current methodologies, benchmarking performance across domains, and identifying emergent trends, this work not only synthesizes existing knowledge but also outlines promising directions for future research at the intersection of AI and social science. Finally, we highlight how these advancements are reshaping societal norms and enabling AI-driven solutions to pressing global issues, from autonomous navigation to public health and beyond.

Through this work, we have proposed a deep CNN architecture based Unet model, to automatically segment thoracic CT scans images. At first, the database has been augmented using preprocessing algorithms such as rotations and filtering. The second step, we have proposed a deep CNN architecture based two models: Unet and Segnet . High performance scores including accuracy, F1-score have been reached. Moreover, the Unet proved to be efficient to segment different regions such as normal lungs, glass representing Covid-19 and background regions. Results quantification have been reached to 97.98% for the train and 95.49% for the validation. The segmented tested images represent an error too close to zero, in comparison to the ground truth. The proposed segmentor system shows to be reliable, achieving an IOU value equal to 92% and 87%, an MSE equal to 0.08 and 0.13 for the validation and test processing respectively. Thus, our proposed system could be a good candidate to help doctors for fast covid-19 diagnosis.

The research is based on the importance of the agentic AI and generative AI that can change the current banking services. The entire analysis has been executed using secondary qualitative analysis using journals and other reports. A total of three themes are generated that are decision-making and automation using agentic AI, personalisation and fraud detection using generative AI, and integration issues, together with future opportunities. The results indicate that these two tools and hybrid human-AI decision-making are used to mitigate fraud-related activity in the banking sector. These findings cover gaps in the past research, which concentrated on isolated applications. The study finds that ethical adoption of agentic and generative AI will lead to innovation, build customer confidence, and deliver sustainable banking solutions.

Artificial Intelligence (AI) has gained prominence in recent years, being widely applied in academic and industrial contexts. Its popularization has raised several challenges, particularly the need to make AI models auditable. Explainable Artificial Intelligence (XAI) seeks to address this issue through methods that interpret the decisions of black-box models. Despite its progress, few studies integrate XAI into the software engineering cycle. At the same time, the European Union's AI Act (Regulation 2024/1689) requires extensive documentation for high-risk systems, often resulting in hundreds of pages of reports. To address this gap, this work proposes An Initial UML Approach Towards EU AI Act Compliance, introducing stereotypes, tagged values, and relationships for LIME-based explanations. By graphically representing critical XAI elements, the approach enhances traceability and auditability while providing partial coverage of AI Act requirements, serving as a structured complement to mandatory textual documentation. The proposal is illustrated through a case study involving a breast cancer diagnosis system.

The use of artificial intelligence (AI) in biomedical imaging and genomics has accelerated the pace of precision medicine, as it enables the automatic analysis of complex and high-dimensional data. Nevertheless, the potential dangers of algorithm bias persist in correlation with these improvements, due to training data imbalances, population heterogeneity, and the technical complexities of imaging or genomic data collection. This type of bias may have discriminatory consequences and undermine the reliability and ethical application of AI in the clinical context. This paper advances a dynamically adaptable explainable AI (XAI) model by providing a way of balancing exploration while having the capacity to dynamically update its decision pathways without losing transparency or interpretability. The model also makes use of multi-modal data fusion approaches, dimensionality reduction prior (e.g., autoencoders and t-SNE), and bias-sensitive optimization layers to maximize generalization
across different populations. Methods to explain explainability entail SHAP, LIME, and Grad-CAM to reveal the model rationale both on genomic markers and on imagery features. We show through publicly available high-dimensional datasets that our approach is effective and helps greatly to improve fairness metrics without loss of predictive ability. This is work that not only fills a crucial knowledge gap in the biomedical AI research but also establishes a basis for the introduction of explainable and ethical decision-support tools in genome-based diagnostics.

Adaptive Explainable AI for Fair & Transparent Biomedical Imaging Genomics
Background: Artificial Intelligence (AI) is transforming biomedical imaging genomics, enabling rapid and automated analysis of complex high-dimensional data. Yet, algorithmic bias—stemming from data imbalances, population heterogeneity, and technical collection artifacts—remains a critical challenge. Such biases may undermine both the reliability and the ethical application of AI in clinical practice.

Our Contribution :
This study introduces a dynamically adaptable Explainable AI (XAI) model designed to mitigate bias while preserving interpretability and predictive power.
Core Features :
Dynamic Adaptability: Decision pathways update continuously without sacrificing transparency.
Multi-Modal Fusion: Integration of imaging and genomic data streams.
Bias-Sensitive Optimization: Custom layers to maximize fairness across diverse populations.
Explainability Methods: SHAP, LIME, and Grad-CAM reveal model reasoning at both genomic and imaging levels.
Dimensionality Reduction: Autoencoders & t-SNE manage high-dimensional complexity.

Results:
Improved fairness metrics without loss of predictive accuracy.
Robust performance on publicly available high-dimensional datasets.
Enhanced trustworthiness and clinical interpretability of biomedical AI systems.

Impact: This work addresses a crucial gap in biomedical AI research, providing a foundation for ethical, transparent, and bias-mitigated decision-support systems in precision medicine and genome-based diagnostics.

Keywords: Adaptive AI • Explainable Artificial Intelligence (XAI) • Biomedical Imaging • Genomics • Bias Mitigation • High-Dimensional Data • Fairness • SHAP • Deep Learning • Multi-Modal Analysis

#BiomedicalAI #Genomics #MedicalImaging #ExplainableAI #BiasMitigation #AdaptiveAI #PrecisionMedicine #FairAI #ResponsibleAI #DeepLearning

Explainable Artificial Intelligence (XAI) addresses one of the most critical challenges in machine learning. That is the opacity of complex models. While traditional AI models offer powerful, predictable capabilities, their lack of interpretability creates obstacles for adoption in high-stakes business applications. This paper will explore the principles, methodologies, and real-world implementation of explainable artificial intelligence in business environments. It focuses on how transparency and interpretability foster trust, better decision-making, and accountability. It draws on current literature. The study systematically examines XAI frameworks and their applications in various industries. These industries include manufacturing, finance, and healthcare. It also discusses emerging trends, challenges, and the path forward for integrating XIA in enterprise-level systems.

This paper introduces TORNADO, a cloudintegrated robotics platform designed to tackle the challenges of autonomous manipulation in dynamic indoor environments, particularly those involving small, soft, or deformable objects. TORNADO integrates large-scale foundation models for perception, language comprehension, and high-level reasoning, to achieve strong zero-shot generalization across a wide range of tasks. At its core, the platform features an adaptive cognitive pipeline capable of dynamically reconfiguring its modules-including semantic 3D SLAM, people-aware navigation, dexterous manipulation, and human-in-the-loop learning-to manage uncertainty and adapt to changing conditions. Additionally, TORNADO incorporates a multi-modal Learning-from-Demonstration interface and an Explainable AI engine, enhancing transparency and easing the burden of teaching new tasks. The system is validated through three industry-relevant scenarios: (1) flexible gear and ply-sheet handling in a mechanical parts factory, (2) patient support in a hospital palliative ward, and (3) product sampling and waste management in a distribution center. TORNADO aims to significantly enhance the agility, safety, and overall task performance of mobile manipulators operating in dynamic, human-centric environments.

Theoretical foundations and definitions  Methodological approaches (inherently interpretable models, post-hoc methods, attention-based techniques)  Technical challenges and limitations  Domain-specific applications (healthcare, finance, criminal justice, education)  Regulatory and ethical frameworks  Current research trends and future directions  Implementation challenges and barriers  Case studies and practical applications  Policy and societal implications  Comprehensive conclusion  Academic references

Artificial Intelligence (AI) multi-agent frameworks are enabling autonomous decision-making, intelligent collaboration, and the automation of complex workflows. These frameworks leverage Large Language Models (LLMs) and distributed AI systems to optimize operations across diverse sectors, with finance emerging as one of the most impacted domains. AI agents are increasingly employed in risk assessment, regulatory compliance, algorithmic trading, fraud detection, and customer service, fundamentally altering how financial institutions operate and manage market dynamics. This paper presents a review of AI multi-agent frameworks, evaluating their architectures, applications, and deployment challenges within financial services. We conduct an in-depth comparative analysis of prominent frameworks, including LangChain, CrewAI, and OpenAI Swarm, assessing their strengths, limitations, and suitability for different financial applications. Furthermore, we examine how these

In a modern world there is a growing intricacy in working places such as shift remote and hybrid working methods as indulge due to this there is need for real time performance monitoring. Traditional methods of performance monitoring files to give us insight into the problem but a modern and machine learning enables the continuous performance assessment through the various tools such as predictive analytical anomaly detection recommendation. this paper examines the current status of ml driven performance tracking and lighten up on strength limitation and its consideration we combine the recent research to identify the recent trends and the future in the practical execution.

Machine learning (ML) is transforming our understanding of health and disease, laying the groundwork for
precision medicine and computational biology. ML algorithms are adept at modeling complex patterns using
heterogeneous and voluminous biological data. This approach represents a significant shift from traditional
reductionism, which seeks to identify single causative variables, to probabilistic models that account for the
interactions, emergence, and uncertainties inherent in biological data about genetic, molecular, and environmental
factors. By capturing contextual dependencies and mitigating noise from isolated experiments, ML facilitates the
creation of more comprehensive representations of biological phenomena. This evolution moves healthcare
toward adaptive, data-driven reasoning at the core of clinical decision-making, diagnosis, and treatment.

This research work looks into the challenges of implementing Explainable Artificial Intelligence (XAI) for threat detection in the banking network. The introduction of XAI is geared towards providing more transparency and accountability in AIdriven decision-making process, which will also address the concerns of reliability and trust in its process. The research also highlights the benefits and prospects of XAI in the banking network and its role in improving threat detection and reducing cases of false alarm in financial transactions. It also further discussed key challenges that impedes the successes of XAI in banking network such as complexity, data quality and compliance among others. The Author went further to provide recommendations and way further in addressing these challenges in implementing XAI in banking networks. The research therefore concluded that the emergence of XAI has the potential to enhance threat detection in the banking networks, but its implementation calls for more careful consideration as XAI will be more significant in banking networks going forward. I.

Credit risk evaluation is central to financial stability and prudent lending practices. Conventional approaches, such as logistic regression, remain widely used because of their simplicity and ease of interpretation, yet they often struggle to detect the nonlinear patterns that characterize borrower behavior. Recent advances in
machine learning (ML) — including algorithms like Random Forest, XGBoost, and Neural Networks provides stronger predictive performance but are frequently criticized for their opacity, which raises concerns for practitioners and regulators.
To address this gap, this study introduces an Explainable Artificial Intelligence (XAI)
framework that integrates the predictive advantages of ML with interpretability methods such as SHAP and LIME. Drawing on an open-source credit dataset, the analysis compares traditional and ML models across multiple evaluation criteria
Results: Results indicate that Logistic Regression delivers the highest overall accuracy (77.3%), while ensemble techniques such as Random Forest show greater effectiveness in distinguishing high-risk borrowers. The XAI tools further reveal that loan amount, borrower age, and loan duration are the most influential factors driving default risk, offering valuable insights for both lenders and policymakers.
Overall, the findings suggest that embedding explainability within ML applications can achieve a balance between predictive precision and transparency, thereby supporting more responsible and trustworthy adoption of AI in credit risk management

Predicting the compressive strength of concrete presents a significant challenge due to its complex composition. Traditional approaches often grapple with data uncertainty, hindering accurate predictions. This study introduces an innovative methodology that integrates advanced preprocessing techniques with a suite of sophisticated prediction models. Employing a robust dataset from the UC Irvine Machine Learning Repository, comprising 1030 samples with eight distinct input variables, the methodology synergizes statistical methods, Bayesian analysis, and cutting-edge machine learning algorithms. Central to our approach is the application of normalization strategies Box-Cox transformation. These predictions were combined with Gaussian Process Regression (GPR), Multiple Linear Regression (MLR), and Support Vector Regression (SVR) to enhance predictive accuracy. The performance of these models is meticulously evaluated using a comprehensive set of metrics, including Root Mean Square
Error (RMSE), R-square, Mean Absolute Error (MAE), and Mean Square Error (MSE). Our findings reveal a marked improvement in prediction accuracy, with GPR emerging as the most effective model. This study not only advances our understanding of concrete's compressive strength but also sets a precedent for employing a multi-faceted analytical approach in material science research.

The biomanufacturing industries are based on the differences in systems, which include Laboratory Information Management Systems (LIMS), Manufacturing Execution Systems (MES), and Quality Management Systems (QMS) causing compromises with transparency, delays in decision-making process, and operational inefficiency in Chemistry, Manufacturing, and Controls (CMC) processes. This paper introduces an interoperability architecture that would cut across these silos and create a decision fabric that incorporates real-time shop-floor signals in high-level portfolio decision support. With the help of Data Fabric and Data Mesh concepts, the suggested solution will allow the data flow to be seamlessly organized, keeping data of LIMS, MES, and QMS systems structured and unstructured to be aligned. The architecture application transports shop-floor sensor information, quality control reports and manufacturing records into a centralized decision layer, enabling analytics-informed insights. The framework is focused on scalability, data governance and security to comply with the industry requirements. The complex data models and metadata-driven methods guarantee the contextual data knowledge and provenance. A case study illustrates how the architecture enhances speed of transparency, less data redundancy and predictive decision-making in CMC operations. Findings suggest that there were immense gains in operational visibility, quicker issue resolution and higher strategic decisions at the portfolio level. The given strategy of integration will lead to the transition of the disjointed data management to the single decision support fabric and enable the stakeholders, giving them access to the timely, correct and practical data. The architecture can be scaled to upcoming digital twin and AI-based optimization plans.

AI decision-making process, especially within CMC systems, it is paramount to consider ethical and explicable
aspects for regulated applications. This study examines the significance of ethics and explainability in AI-driven
CMC decisions by assessing governance frameworks that promote model validation, bias control, and auditability.
We look at how XAI methods increase user confidence, enhance decision transparency, and ensure regulatory
compliance. The research leverages recent advancements in ethical AI models, bias mitigation techniques, and
governance frameworks to ensure the safe implementation of AI systems within cybersecurity, healthcare, and
financial risk analysis domains. Additionally, it touches on the challenges of introspection, interpretability, and
bias elimination in opaque models as well as related regulatory issues and best practices to combat biases.
Ultimately, there is a need to pay attention to the ethics and explainability of AI as design features to create
trustworthy CMC.

The integration of artificial intelligence (AI) into financial statement analysis has marked a significant shift in how financial data is processed, analyzed, and interpreted. This paper provides an in-depth examination of the impact of AI on financial statement analysis, highlighting its benefits, limitations, and future directions. We explore the current state of AI applications in financial analysis, including machine learning, natural language processing, and deep learning, and discuss their implications for financial statement analysis. *Introduction* Financial statement analysis is a critical component of accounting and finance, enabling stakeholders to assess a company's financial health, performance, and prospects. Traditional methods of financial statement analysis rely heavily on manual processing and analysis of financial data, which can be time-consuming, prone to errors, and limited in scope. The emergence of AI has transformed the field of financial statement analysis, offering new opportunities for automation, insight generation, and decision-making. *Benefits of AI in Financial Statement Analysis* 1. *Improved Accuracy*: AI algorithms can process large datasets with high accuracy, reducing errors and inconsistencies associated with manual analysis. For instance, AI-powered tools can accurately extract financial data from complex financial statements, reducing the risk of human error.

Early Autism Spectrum Disorder (ASD) detection is important for early intervention. This study investigates the potential of eye-tracking (ET) data combined with machine learning (ML) models to classify ASD and Typically Developed (TD) children. Using a publicly available dataset, five ML models were evaluated: Support Vector Machine (SVM), Random Forest, Convolutional Neural Network (CNN), Artificial Neural Network (ANN), and Random Forest improved with Convolutional Filters (ConvRF). The models were trained and tested using a set of evaluation metrics, including accuracy, precision, recall, F1-score, and ROC Area Under the Curve (AUC). Among these, the ConvRF model attained superior performance, achieving a recall of 90% and an AUC of 88%, indicating its robustness in identifying ASD children. These results highlight the model’s effectiveness in ensuring high sensitivity, which is critical for early ASD detection. This study shows the promise of combining ML and eye-tracking technology as accessible non-invasive tools for enhancing early ASD detection, resulting in timely and personalized interventions. Limitations and recommendations for future research are also included.

While Deep Neural Networks (DNNs) can perform a wide range of tasks at human or greater-than-human level of competence, they are also notoriously opaque. This paper aims to shed light on both the specific nature of this opacity and what it would take to fully or partially remove it. We begin by drawing a clarificatory distinction between two basic dimensions of opacity of complex systems-internal and relational-and explain how various kinds of opacity invoked in recent discussions of DNNs can be understood in terms of these basic dimensions. We then discuss the prospects of removing opacity from DNNs following the methods of two subfields of research in AI-mechanistic interpretability and explainable artificial intelligence (XAI). We finish with a critical discussion of the relevance of what Sullivan (2022a) calls 'link uncertainty'.

Students are exposed to technology at an early age, yet the rapid expansion of digitalisation outpaces the development of policies and frameworks to guide their interactions with it. Despite its relevance, digital ethics literacy remains largely unexplored, creating risks and challenges for younger children who often lack awareness of responsible and safe digital practices. This study aims to promote digital ethics literacy by implementing and analysing three lessons, carried out by the school counselling, in two fourth grade classes at a German primary school. It seeks to investigate the extent of awareness among fourth grade students regarding the risks of digital practices, to explore how participative, dialogical, problem-posing, critical and context-based learning approaches can improve students' understanding of digital ethics, and to identify strategies that primary schools can implement to foster ethical digital engagement. To do so, this research adopts a child-centred approach, conducting an ethnographic study involving thirty-three students. The research combined participant observation, informal inquiries, field notes and a student lesson evaluation designed to analyse the learning experience from the children's perspective. The collected data were analysed to describe and interpret students' interactions and discussions in the classroom, to evaluate their engagement with digital environments and the impact of the applied teaching methods, as well as to make suggestions on how to promote digital ethics in primary schools. The results indicate a high level of digital engagement among students but a limited awareness of risks, such as communicating with strangers, sharing personal information and access to manipulative or inappropriate content, including hate speech. Digital practices were often associated with negative emotions and conflicts. Nevertheless, students-particularly those receiving parental guidance-were able to propose solutions to mitigate challenges. The applied learning approach to digital ethics literacy fostered active participation, which contributed to high levels of engagement.

You are invited to submit chapter proposals for a planned edited Volume with Routledge entitled Higher Education, AI, and Ethics: Problems & Prospects for Student Flourishing in the Academy. This edited collection of chapter brings together critical and diverse perspectives on the profound ethical, pedagogical, and societal questions arising from the rapid adoption of Artificial Intelligence (AI) in Higher Education (HE). Context and Rationale Universities face unprecedented political, economic, and social pressures, with their role as institutions for the public good increasingly under strain. Among the most urgent challenges is the integration of AI into teaching, learning, and research. From questions of authorship and authenticity to wider societal concerns about trust, equity, and democracy, the implications of AI demand collective and global attention. This volume addresses pressing concerns at the intersection of AI, ethics, and higher education, including: • AI's impact on authorship, intellectual labour, and academic integrity. • Transparency, accountability, and the ethics of AI-assisted learning. • Challenges of detection in evaluation processes and the erosion of trust in the teacher-student relationship. • Fragmented and inadequate institutional responses. • Broader social dangers, including diminished critical thinking, weakened democracy, and reinforcement of inequalities. While much discourse on AI in HE is either optimistic (policy-driven, innovationfocused) or sceptical (philosophical critique), there has been little sustained, integrative attention as to how AI reshapes conditions for student flourishing and academic life. This volume seeks to fill that gap.

Machine Learning (ML) is already part of everyday life, powering applications such as Netflix recommendations, Google Maps navigation, and online fraud detection. This article surveys ten practical, real-world use cases of ML, explaining both the technical models and workflows behind them. Case studies include Netflix’s recommendation engine, Google Assistant’s speech models, and Tesla’s autonomous driving systems. By combining technical insights with practical examples, this study highlights how ML is shaping modern society while pointing to future directions such as emotional AI and real-time translation.both researchers and practitioners.

In order to meet the explainability requirement of AI using Deep Learning (DL), this paper explores the contributions and feasibility of a process designed to create ontologically explainable classifiers while using domain ontologies. The approach is illustrated with the help of the Pizzas ontology that is used to create a synthetic image classifier that is able to provide visual explanations concerning a selection of ontological features. The approach is implemented by completing a DL model with ontological tensors that are generated from the ontology expressed in Description Logic.

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

Air travel during pregnancy poses unique physiological risks, including deep vein thrombosis (DVT), radiation exposure, and cabin pressure effects on fetal oxygenation, necessitating rigorous fitness-to-fly assessments [?]. Current guidelines recommend medical clearance after 28 weeks of gestation or in cases of complications,

Concrete is one of the building industry's most used construction materials. Reducing natural resources, enormous production costs, and environmental issues in cement production have encouraged researchers to partially explore suitable options to substitute Portland cement and the built environment. This article provides a thorough review of the research on the prudent utilization of sugarcane bagasse ash (SCBA) as a Portland cement replacement in the production of concrete. The methods used to produce SCBA, the effect of calcining temperature on bagasse ash, its physical and chemical characteristics and the strength development phenomenon are discussed. The impact of SCBA on the properties of concrete under the fresh state is also discussed. The physical and durability properties of concrete manufactured with SCBA are reviewed in-depth to understand its impending use for commercial applications. Finally, SCBA-related issues and challenges are described. A few of the outcomes are: (i) An organized incineration method is required for producing good quality SCBA, (ii) the optimum replacement level of cement by SCBA for mechanical and durability properties are 20%, (iii) improved durability due to an impervious microstructure of SCBA-concrete to harmful agents that cause degradation. Even though there is some disagreement among researchers, the majority continue to agree that using SCBA in cementitious composites is advantageous. However, researchers do not consider its usage in reinforced concrete elements such as slabs and beams; thus, further research is recommended. Finally, the formulation of codal recommendations on technical and environmental factors calls for additional research.

Investigations on the potential use of sustainable sugarcane bagasse ash (SCBA) as a supplementary cementitious material (SCM) in concrete production have been carried out. The paper employs model agnostic eXplainable Artificial Intelligence (XAI) to develop interpretable models for the compressive strength of SCBA concrete. A dataset of SCBA concrete, comprising of 2616 data points, was first established. Then, black box machine learning models were employed for predicting compressive strength of SCBA concrete. White box modelling using model agnostic XAI explanations for the influence of different input parameters on the compressive strength of SCBA concrete was then used for the local and global interpretations of influencing parameters. Based on the relative influence of features on the compressive strength of SCBA concrete, a nonlinear model equation was thus developed. Optimization of SCBA concrete mixes was done and it was found that a compressive strength of 40 MPa could be achieved for water to binder content ranging from 0.42 to 0.48 with low cement content. Additionally, the Life-Cycle Assessment (LCA) was carried out, and finally, it was proposed that SCBA concrete has potential for becoming an alternate eco-friendly building material.

JEL Classifications O33-Technological Change: Choices and Consequences; Diffusion Processes (Primary): This code fits the article's core discussion of AI technologies (e.g., aligned/unaligned systems, multi-agent dialogue) as tools for historical forensic auditing, including their implications for bias detection and research efficiency. It encompasses studies on AI's societal and methodological impacts, such as innovation in document verification and historical interpretation. B41-Economic Methodology: Appropriate for the methodological framework of using AI dialogue to audit history, including bias mitigation and verification processes. The article critiques traditional vs. AI-enhanced approaches, aligning with JEL's emphasis on research methods and tools in knowledge production. D83-Search; Learning; Information and Knowledge; Communication; Belief; Unawareness: This covers the article's exploration of AI in information processing, fact-checking, and bias in historical data. It relates to how AI systems handle uncertainty, learning from sources, and collaborative dialogue in knowledge generation. C63-Computational Techniques; Simulation Modeling: Relevant for the technical aspects of multiagent AI systems, adversarial verification, and simulation in historical analysis, including flowchartbased processes described in the article. N01-Development of the Discipline: Historiographical; Sources and Methods (Secondary, if emphasizing economic history): This could apply if framing the article within economic historiography, as it discusses auditing historical narratives through AI, though the content is more broadly technological than purely economic.

Telemedicine has become a critical enabler of healthcare delivery, particularly in the post-pandemic era, facilitating crossborder consultations, remote monitoring, and mobile health (mHealth) services. However, these innovations raise substantial concerns about patient data security, privacy, and compliance with varying international data protection regulations. This article investigates how privacy-preserving technologies (PPTs), including blockchain, homomorphic encryption, and federated learning, can strengthen trust, ensure compliance, and mitigate risks in telemedicine platforms operating across diverse regulatory environments. We conducted a systematic scoping review of peer-reviewed literature published between 2015 and 2025 across databases including PubMed, IEEE Xplore, Scopus, and Web of Science. Additionally, a comparative policy analysis of international data protection frameworks (GDPR, HIPAA, NDPA 2023, and Asia-Pacific regulations) was carried out, complemented by expert interviews with telehealth providers in Europe, Africa, and North America.

This paper presents an adaptive multi-model framework for cybercrime identification and prediction by integrating machine learning with explainable artificial intelligence (XAI). A multi-stage pipeline is developed that preprocesses cybercrime-related text, applies advanced ML models for classification, and incorporates XAI techniques such as LIME and SHAP to enhance interpretability. The framework not only achieves high accuracy in detecting malicious communication but also provides human-understandable justifications for each prediction, thereby improving trust and accountability. With realtime monitoring and continuous learning capabilities, the system is designed to evolve with emerging cybercrime patterns, ensuring robustness and applicability in diverse domains such as social media moderation, enterprise communication security, and law enforcement support.

As artificial intelligence transforms industries at breakneck speed, we're witnessing a fundamental debate about how to implement these powerful tools. This paper explores the growing tension between human-centric and tech-centric approaches to AI, making the case that artificial intelligence works best when it amplifies human capabilities rather than substituting for them. We draw on Garry Kasparov's groundbreaking insights about human-machine collaboration in chess, the U.S. military's innovative digital centaur framework, and cutting-edge research from leading institutions to outline practical strategies for keeping AI tools under human direction while unleashing their full potential to improve decision-making across industries. Through detailed analysis of spectacular AI failures-from legal practitioners facing hefty sanctions to discriminatory lending algorithms-we examine what happens when organisations rely too heavily on automated systems. The paper also tackles the thorny issues of AI-generated text summaries and the cognitive effects of AI dependency. We wrap up by exploring what's really at stake when organisations get the human-AI balance wrong.

Human-induced global warming, primarily attributed to the rise in atmospheric CO 2 , poses a substantial risk to the survival of humanity. While most research focuses on predicting annual CO 2 emissions, which are crucial for setting long-term emission mitigation targets, the precise prediction of daily CO 2 emissions is equally vital for setting short-term targets. This study examines the performance of 14 models in predicting daily CO 2 emissions data from 1/1/2022 to 30/9/2023 across the top four polluting regions (China, India, the USA, and the EU27&UK). The 14 models used in the study include four statistical models (ARMA, ARIMA, SARMA, and SARIMA), three machine learning models (support vector machine (SVM), random forest (RF), and gradient boosting (GB)), and seven deep learning models (artificial neural network (ANN), recurrent neural network variations such as gated recurrent unit (GRU), long short-term memory (LSTM), bidirectional-LSTM (BILSTM), and three hybrid combinations of CNN-RNN). Performance evaluation employs four metrics (R 2 , MAE, RMSE, and MAPE). The results show that the machine learning (ML) and deep learning (DL) models, with higher R 2 (0.714-0.932) and lower RMSE (0.480-0.247) values, respectively, outperformed the statistical model, which had R 2 (-0.060-0.719) and RMSE (1.695-0.537) values, in predicting daily CO 2 emissions across all four regions. The performance of the ML and DL models was further enhanced by differencing, a technique that improves accuracy by ensuring stationarity and creating additional features and patterns from which the model can learn. Additionally, applying ensemble techniques such as bagging and voting improved the performance of the ML models by approximately 9.6%, whereas hybrid combinations of CNN-RNN enhanced the performance of the RNN models. In summary, the performance of both the ML and DL models was relatively similar. However, due to the high computational requirements associated with DL models, the recommended models for daily CO 2 emission prediction are ML models using the ensemble technique of voting and bagging. This model can assist in accurately forecasting daily emissions, aiding authorities in setting targets for CO 2 emission reduction. Keywords Daily CO 2 emissions • Prediction and forecast • Machine learning model • Deep learning model • Statistical model

The emergence of intelligent medical diagnostic systems fuelled by the pandemic is primarily focused on imaging data. While Convolutional Neural Networks (CNNs) are powerful tools in extracting visual features, their integration with neuro-fuzzy inference systems and dimensionality reduction methods like PCA enhance interpretability and performance on a broader scale. This work proposes a hybrid diagnostic framework for chest CT scan images that incorporates CNN for feature extraction together with PAS and ANFIS for/classification to accurately detect COVID-19. The system performs component extraction through CNN, reduces dimensionality by P mon, and uses ANIFIS to make the final class designation. The hybrid model is evaluated on a recent and thoroughly annotated dataset where it has proven to outperform the previous models in accuracy, robustness, ease of use and interpretative evaluation. This highlights the potential of the model for real-life application.

Cloud security has emerged as one of the most challenging issues in the field of information technology. This paper presents a novel intelligent framework to counter security threats in cloud environments by combining advanced machine learning techniques with domain expert knowledge. The proposed system, leveraging Natural Language Processing (NLP) and Large Language Models (LLMs), is capable of automatically identifying and classifying threats. Key findings indicate that the proposed method achieves a detection accuracy of 94.2%, representing a 22.4% improvement over conventional approaches. Notable features of the system include an 85-millisecond response time, a 49% reduction in resource consumption, and adaptive learning capabilities. Real-world deployment results demonstrate a 62% decrease in successful attacks and an annual cost saving of $250,000. The main advantage of this framework, compared to existing solutions, lies in its intelligent integration of three sources of knowledge: Structured data from a Systematic Literature Review (SLR), Domain expert insights, and Advanced machine learning models. This combination enables the system to deliver both high analytical accuracy and interpretable decision-making. This study opens new pathways for future research in advanced cybersecurity and offers practical applications that could significantly transform the cloud computing industry. The results are particularly valuable for organizations and cloud service providers facing complex security challenges.

W pracy rozważam, w jaki sposób konstytucyjne prawo dostępu do informacji publicznej (art. 61 Konstytucji) prowadzi do konieczności transparentnego informowania o technologiach wspomagających władzę publiczną. Wyróżniam potrzebę unikania „instytucjonalnych czarnych skrzynek” – czyli zapewniania dostatecznej jawności informacji na temat wykorzystywanych technologii wspomagania władzy – oraz potrzebę unikania „algorytmicznych czarnych skrzynek” – objawiającą się w potrzebie preferowania takich algorytmów, na temat których dane mogą być rzeczywiście zrozumiałe dla obywateli. Zarazem charakteryzuję wymóg instytucjonalnej transparentności jako istotniejszy w tym sensie, że podlegający mniejszej liczbie wyjątków. Z kolei wymóg algorytmicznej transparentności zdaje się bardziej stopniowalny i zależny od pola zastosowania danej technologii. Zarazem wykorzystanie algorytmicznych czarnych skrzynek tworzy ryzyko fasadowego charakteru udostępniania informacji na temat takich nietransparentnych algorytmicznie technologii. Dlatego w tych dziedzinach działalności państwa, w których oczekiwany jest wysoki poziom przejrzystości dla obywateli, art. 61 Konstytucji zdaje się tworzyć po stronie władzy publicznej wymóg zarówno informowania o wykorzystywanych technologiach, jak i właściwego ich projektowania.

Phishing is a common form of cyberattack where fraudulent websites are used to deceive users into revealing sensitive information. Detecting phishing sites is crucial for enhancing online security. This study presents a machine learning-based system for phishing detection using URL features. A balanced dataset containing phishing and legitimate samples was used to train and evaluate the models. The system extracts 23 essential features from URLs and employs machine learning models such as Random Forest, SVM, Gradient Boosting, and MLP for classification. Among these, the Random Forest model achieved the highest accuracy of 96.17%. The system is deployed as a web application using Flask, providing real-time detection and easy user interaction. This research highlights the effectiveness of machine learning in improving cybersecurity through accurate and efficient phishing detection.

The escalating complexity and widespread deployment of autonomous systems, ranging from advanced industrial robotics to intelligent urban infrastructure, necessitate a paradigm shift in software engineering. These systems demand not only high adaptability but also rigorous security and transparent decision-making. This paper proposes a unified Software Intelligence Framework that seamlessly integrates Neural Program Synthesis (NPS), Quantum-Secure DevOps (QSD), and Explainable AI (XAI) to meet these multifaceted demands. The framework leverages advancements in NPS for AI-driven code generation, QSD for fortifying software lifecycles against emerging quantum threats, and XAI for ensuring interpretable and trustworthy decisionmaking in critical autonomous operations. We present a comprehensive literature review of the state-of-the-art in each domain, detailing their respective challenges and synergistic potential. The proposed architecture unifies these components into a continuous pipeline for specification-to-code generation, secure deployment, and runtime adaptation. A hypothetical smart city infrastructure scenario illustrates the practical application and benefits of this convergent framework, demonstrating its capacity for rapid code adaptation, post-quantum security, and human-understandable explanations of autonomous behavior. We further discuss the technical challenges inherent in such integration, along with robust evaluation strategies and the profound ethical, operational, and security implications of deploying AI-generated, quantum-secure systems in sensitive contexts. This work lays the foundation for a new multidisciplinary field essential for developing adaptable, robust, and trustworthy autonomous systems.

Breast cancer is the predominantly diagnosed cancer malignancy among women globally with a massive significant percentage and sizeable rate of cancer-related morbidity and mortality, annually in the United States. Breast cancer genetic testing and metastasis analyzes genetic markers in individuals with cancer to predict their future susceptibility to cancer and is an inherent part of early detection, prevention, treatment, and intervention of cancer. Immeasurable studies evince the monumental significance of Artificial Intelligence (AI) in clinical medicine, scientific research innovation and revolutionary patient care, treatment and decision-making processes in breast cancer genetic testing and metastasis. Notably, Explainable Artificial intelligence (XAI) develops AI models and algorithms to facilitate understandable complex cancer genetic data through explanations for AI's predictions and reasoning to identify relevant genetic mutations, pathways, and interactions that contribute to cancer development and progression. This comprehensive overview delves into the egregious advances of XAI in breast cancer genetic testing and metastasis, in comprehending AI decision-making processes of various AL algorithms with model explainability, monitoring, auditability, accountability, compliance as well as other risks factors associated with executing trustworthy and accurate AI in the screening and management of breast cancer. The findings of this overview corroborate and evaluate substantial perspectives of XAI in breast cancer genetic testing and metastasis.

Background. Accuracy alone does not capture the pedagogical value of human-AI dialogue. We introduce Contextual Loops (CTL)-a mechanism-level lens on conversational dynamics-and an ordinal Semantic Progression (SP, 0-3) rubric that distinguishes fluent restatement from genuine advances in reasoning. To our knowledge, this preprint is the first formalization of CTL and SP within a controlled Hybrid-Lab protocol designed for short, classroom-and workflow-realistic exchanges. Objective. To estimate whether a newer model more reliably advances Semantic Progression and reduces stagnation-prone dynamics than a strong prior baseline, and to quantify the reliability of the rubric and raters. Methods. We conducted 45 prompt-matched trials spanning six prompt archetypes (PA-01…PA-06). Generation and rating roles were separated (generator ≠ rater); raters were blinded and included a human anchor plus calibrated AI raters. The primary analysis used a cumulative-link mixed model (proportional odds) with fixed effects for Model (GPT-5 vs GPT-4o), RaterType (AI vs human), Context (ON/OFF), and Tools (ON/OFF), and random intercepts for conversation and rater. Secondary analyses assessed saturation risk and CTL⁺ dominance; survival curves characterized time-to-drift onset. Reliability used Krippendorff's α and quadratic-weighted κ. Results. GPT-5 showed higher odds of advancing SP than GPT-4o (POR = 2.05, 95% CI [1.21, 3.47], p = 0.007), corresponding to marginal P(SP≥2) of 0.61 vs 0.38 (absolute Δ = +0.23). Relative to PA-01 (Validation-Seeking), PA-02 (Moral/Ethical) (POR = 1.84 [1.05, 3.22], p = 0.034) and PA-03 (Contradiction) (POR = 1.96 [1.10, 3.49], p = 0.022) were higher, while PA-06 (Echo Reinforcement) trended lower (POR = 0.71 [0.41, 1.20], p = 0.200). RaterType effects were small and non-significant (POR = 1.12 [0.89, 1.41], p = 0.330); Context and Tools were directionally positive and null, respectively (Context POR = 1.18 [0.93, 1.49], p = 0.160; Tools POR = 1.05 [0.82, 1.36], p = 0.680). Saturation risk was lower for GPT-5 (OR = 0.58 [0.36, 0.92], p = 0.021), and CTL⁺ odds were higher (OR = 2.34 [1.20, 4.60], p = 0.013). Reliability was substantial (Krippendorff's α = 0.72 [0.64, 0.79]), with human↔AI κ = 0.66, 0.61, 0.68. Survival curves indicated a later drift onset for GPT-5, consistent with reduced stagnation within short windows. Conclusions. Within brief, authentic exchanges, GPT-5 more consistently advances Semantic Progression and resists stagnation than GPT-4o, even when generation and rating are separated, and raters are blinded. These mechanism-level gains clarify how learning value can emerge from dialogue beyond simply "getting it right," and they motivate practical safeguards for high-stakes use: triangulation across models or methods, role separation (generator ≠ rater) to avoid self-grading artifacts, and explicit uncertainty disclosure. Findings are proof-of-concept and limited by scale, scope, and the absence of downstream outcome measures; nonetheless, they provide an empirical foundation for CTL/SP as actionable diagnostics for education, business intelligence, governance and ethics, and healthcare workflows.

chronic kidney disease (CKD) is often diagnosed at later stages, leading to severe health impacts. This study presents a machine learning-based approach for early CKD prediction using patient clinical data. To improve model transparency, Explainable AI (XAI) techniques like LIME are employed, offering insights into feature contributions. The proposed system achieves high accuracy and supports clinical decision-making by identifying key indicators influencing CKD onset.

This paper presents a novel federated learning framework for implementing voice biometric authentication in multi-tenant cloud contact centers while preserving customer privacy and regulatory compliance. Traditional centralized voice biometric systems face significant challenges in cloud environments due to data sovereignty requirements, privacy regulations (GDPR, CCPA), and tenant isolation needs. Our proposed framework enables distributed learning across tenant boundaries without exposing raw voice data, achieving 97.3% authentication accuracy while maintaining strict data locality. The system leverages differential privacy mechanisms and homomorphic encryption to protect individual voice patterns during model aggregation. Implementation results from a Cisco Webex Contact Center environment serving 50+ enterprise tenants demonstrate 40% reduction in authentication latency and 99.8% compliance with data residency requirements. This approach addresses critical gaps in secure biometric deployment for cloud-native contact centers while enabling cross-tenant learning benefits.

Breast cancer remains one of the leading causes of mortality among women worldwide, where early detection and accurate classification are pivotal for effective treatment and improved survival rates. This research presents a novel Convolutional Neural Network (CNN)-based deep learning framework for the automated detection and multi-class classification of breast cancer using mammogram images. The proposed architecture is designed to extract hierarchical features from the input mammograms and classify them into normal, benign, and malignant categories with high precision.The system is trained and evaluated on the publicly available Digital Database for Screening Mammography (DDSM), which provides high-quality grayscale mammogram images. A comprehensive preprocessing pipeline including image resizing, normalization, and augmentation was implemented to enhance model generalization. The CNN architecture comprises three convolutional blocks followed by

This study explores the transformative role of digital innovations-specifically artificial intelligence (AI), the Internet of Things (IoT), and big data analytics-in addressing critical challenges within the Syrian healthcare system, a context defined by limited resources, disrupted infrastructure, and significant healthcare disparities. By integrating these technologies, healthcare delivery can be revolutionized through predictive analytics, remote monitoring, and personalized treatment strategies, thereby enhancing clinical outcomes, operational efficiency, and system resilience. The research investigates the current landscape of digital health adoption in Syria, identifying infrastructural, technical, and policy-related barriers while proposing data-driven frameworks for strategic implementation. Emphasis is placed on how AI-driven diagnostics, IoT-enabled remote care, and analytics-informed decision-making can collectively support evidence-based practices and facilitate the transition towards a patient-centric, precision medicine paradigm. This paper offers targeted insights for policymakers, healthcare providers, and technology developers aiming to modernize healthcare in conflict-affected and resource-constrained environments. The findings contribute to the broader discourse on sustainable healthcare transformation, demonstrating the potential of digital technologies to strengthen healthcare infrastructure, foster health equity, and support long-term development goals. Moreover, the paper underscores the necessity of robust data governance, ethical AI frameworks, and cross-sectoral collaboration to ensure equitable and secure deployment of digital health solutions. Ultimately, this research provides a strategic roadmap for harnessing digital innovation to achieve improved healthcare outcomes and resilience in Syria and similar contexts.

Retinopathy of Prematurity (ROP) is a vision-threatening condition in premature infants requiring timely and accurate diagnosis to prevent blindness. While electronic health (eHealth) technologies promise to improve neonatal care, automating ROP diagnosis faces challenges such as limited labeled datasets, architectural complexity, and high computational demands. This study introduces LightEyeNet, a lightweight deep-learning architecture optimized for eHealth applications in ROP severity classification. By integrating DenseNet121 and a channel-wise residual attention network block, LightEyeNet enhances diagnostic accuracy and efficiency. Explainable AI techniques, including Grad-CAM and LIME, further improve transparency and clinical interpretability. LightEyeNet achieves 96.28% testing accuracy, outperforming state-of-the-art pre-trained networks, including DenseNet201 (95.78%, + 0.5%), Inception-V3 (93.80%, + 2.48%), Xception (94.54%, + 1.74%), and EfficientDense (87.10%, + 9.18%). Furthermore, LightEyeNet is the most compact architecture among these, with a size of 2.45 MB, compared to Efficient-Dense (6.88 MB), Inception-V3 (3.56 MB), Xception (21.48 MB), and DenseNet201 (5.05 MB). With a specificity of 0.99, a sensitivity of 0.95, and an AUC score of 0.99 across five ROP severity classes, LightEyeNet demonstrates a balance of superior performance and efficiency.

Accurate rainfall forecasting underpins effective water resource management, disaster mitigation, and agricultural planning. This paper proposes RfGANNNet 2.0, a hybrid AI framework that combines Random Forests, Spatio-Temporal Graph Convolutional Networks (ST-GCN), and Physics-Guided Generative Adversarial Networks (GANs) to deliver high-resolution and generalizable rainfall predictions. We present a comprehensive review of AI-driven techniques from 2018 to 2025, including ensemble models such as AdaNAS, deep learning architectures like ConvLSTM and Temporal Fusion Transformer (TFT), and spatio-temporal attention mechanisms. The model integrates satellite and radar remote sensing data, addresses limitations related to data sparsity, and incorporates Explainable AI methods (e.g., SHAP, LIME) to interpret model outputs. Extensive evaluations using benchmark datasets and metrics such as RMSE, MAE, and AUC highlight the robustness and accuracy of the proposed approach. Future research directions include real-time edge computing, adaptive transfer learning, and advanced data fusion to improve operational readiness and performance in extreme weather scenarios.