ABSTRACT. As Information Technology (IT) and Operational Technology (OT) environments converge, Industrial Control Systems (ICS) become highly susceptible to advanced cyber threats. Traditional Network Intrusion Detection Systems (NIDS) relying on L3/L4 signature analysis, fail against semantic attacks concealed within legitimate traffic and suffer from high False Positive Rates (FPR) due to natural process noise. To address these limitations, this paper proposes a data-driven fuzzy feature hybridization (DD-FFH) framework integrated with eXtreme Gradient Boosting (XGBoost) to decouple malicious anomalies from natural operational fluctuations in Modbus TCP networks. By injecting expert domain knowledge directly into the feature space via fuzzy logic, the model effectively mitigates the semantic overlap of volumetric and temporal attacks. Experimental validation on a custom IEC 62443-compliant testbed demonstrates that the proposed architecture achieves a 1.0 overall accuracy and a macro-average F1-score of 0.98, successfully classifying attacks. Furthermore, cross-dataset validation on a public ICS dataset provides a broader scope and demonstrates a promising direction for future research in autonomous defense systems.

ABSTRACT. Classifying natural language requirements (NLRs) is a challenging task. Recent studies show that Large Language Models (LLMs) can support automated categorisation of requirements; however, limited research has specifically addressed the distinction between Security Requirements (SRs) and Non-Security Requirements (NSRs). In this work, we investigate the reliability of Generative Pretrained Transformer (GPT)-like models, such as GPT-5.1, when classifying NLRs into SRs and NSRs by using a Zero-Shot Learning (ZSL) approach. Moreover, we explore how different industrial domains can affect the reliability of classification results. We evaluate the results by using standard machine learning evaluation metrics: F1-score (F1), Precision (P), and Recall (R). The model’s reliability varies considerably across application domains, achieving near-human performance in financial systems, moderate performance in security-standard domains, and markedly poorer performance in network-infrastructure domains. The findings highlight the need for caution in practice and the importance of developing more robust, domain-aware approaches.

ABSTRACT. Assessing user experience (UX) and product desirability in video games remains a complex and resource-intensive process, particularly during early stages of development. This paper proposes a lightweight, low-cost evaluation method based on Game User Experience Cards (GUXC) and an aggregation algorithm called GUXC Score. The method adapts the concept of Product Reaction Cards to the video game domain and introduces a quantitative aggregation mechanism that transforms qualitative card selections into a normalized numerical score. A dataset of 693,055 Steam reviews was utilized to construct a domain-specific card set, followed by evaluation with 174 participants (65 in sentiment validation and 109 in main local and remote study). Results demonstrate strong correlations (Pearson r = 0.94 at best) between the GUXC Score and established UX metrics, supporting the method’s validity as an efficient alternative to traditional UX studies.

ABSTRACT. Multi-object tracking in broadcast soccer is difficult because occlusions, scale changes, and crowded interactions frequently disrupt identity continuity. This paper proposes a detector-anchored SAM2 policy that periodically re-aligns propagated tracks with detector outputs to improve trajectory consistency. We benchmark six detector-tracker combinations on SoccerNet-Tracking by pairing two RF-DETR variants with ByteTrack, BoT-SORT, and SAM2, all evaluated under a common protocol. On the full 49-sequence test split, RF-DETR Base + detector-anchored SAM2 achieves the best HOTA (0.512) and IDF1 (0.614) among the tested variants, while classical trackers retain substantially higher throughput. The results show that detector-anchored segmentation-assisted tracking is a strong option for offline tactical analysis where trajectory continuity matters more than runtime.

ABSTRACT. Deep-learning pipelines for sonar-based UXO classification increasingly rely on synthetic data generated in Digital Twin environments. While such pipelines support scalable development of AI-enabled decision-support services, they also risk introducing hidden information leakage, causing models to learn non-physical shortcuts instead of target-relevant acoustic cues. In this paper, we consider a binary classification task (UXO vs.\ non-UXO) and present a five-stage protocol for auditing the integrity of synthetic sonar datasets before further model development and deployment. The protocol combines complexity stress-testing, saliency/Grad-CAM inspection, object masking, and bias-only training on masked images. In our case study, it revealed contamination linked to the simulator data-generation procedure. A leaky dataset achieved near-perfect balanced accuracy (up to 99.22\%), whereas the corrected dataset produced lower but more plausible results (97.35\% on an idealized seabed and 83.62\% on a realistic seabed). The protocol can therefore serve as a practical quality-assurance component for synthetic-data pipelines.

ABSTRACT. This paper proposes a structured, integrative, and auditable methodology for managing missing values in decision support systems (DSS) to improve forecasting reliability. It advocates a shift away from black-box imputation toward a structured workflow that integrates data preprocessing, duplicate detection, multi-faceted missingness analysis, multicriteria selection of imputation strategies, and hold-out evaluation using the same forecast metric optimized by the DSS. The workflow is validated through an extensive study of 172 residential electricity consumption time series. The results demonstrate that formal missingness analysis is essential, as non-random missing patterns identified in all tested series preclude the use of biased, simplistic models. On this single dataset, simpler task-aligned imputation methods outperformed Stable Diffusion 2 generative inpainting by more than 25 percentage points in symmetric mean absolute percentage error (sMAPE), which is a cautionary observation rather than a general recommendation. By providing a clear and transparent path from raw data to validated forecasts, the proposed approach might enhance DSS robustness to missing data.

ABSTRACT. The goal of the research is the analysis of the usability of unsupervised anomaly detection (AD) methods for detection of valve plate failures in piston pumps. The unsupervised AD methods play a special role in predictive maintenance since they do not require failure data to be present during training, which represents a typical scenario of real-world manufacturing operations where faulty states are rarely recorded. A comprehensive comparative analysis of nine detection algorithms, including deep learning models, was conducted using real-world process data. The results assess the models' effectiveness and resilience to industrial noise, identifying optimal solutions for increasing the reliability of monitoring systems in the era of Industry 4.0. The best average result AUC=0.9150 was obtained by the AutoEncoder. The results show that the Isolation Forest model significantly differs from Local Outlier Factor and One-Class SVM in terms of feature importance. Four techniques stand out for their results.

ABSTRACT. The digital transformation is not only a necessity for preserving prosperity in the face of international competition, but also introduces challenges regarding fairness, trust, and bias the more it permeates private areas of life such as finance. This paper examines AI-driven applications in the credit and banking sectors to provide robust methodological guidelines for system development based on the design science research (DSR) paradigm as foundation for creating scientifically sound artifacts that improve the performance of real-world organizations. The findings of a systematic literature review suggest that the interpretability of input data and results is considered more important than understanding the exact mathematical functioning, and that the involvement of human experts is advisable, although they are hardly ever immune to bias. By fostering trust in the responsible use of AI, higher willingness to share personal data could be achieved, which in turn could improve accuracy and the level of transparency.

ABSTRACT. Interoperability remains a major challenge in citizen-centred digital government, particularly when services that should be experienced as a single life event are delivered through disconnected institutional procedures. This paper examines that challenge through the case of newborn registration in Mozambique and frames it as a model-based information systems development problem. Using a design science research approach, we employ DEMO to diagnose the existing coordination structure and then develop a lightweight ArchiMate-oriented artefact to support service redesign. The main result is a DEMO-to-ArchiMate modelling pipeline that links diagnosed coordination failures to an architectural representation of interoperability in a life-event service. From the case, four design requirements are derived and operationalised through five constructs: inter-agency life-event service, interoperability agreement, authoritative semantic asset, orchestration capability, and reusable integration pattern. As an exploratory design-science study, the paper evaluates representational adequacy for redesign, not implementation effectiveness or organisational adoption.

ABSTRACT. As digital transformation increasingly reshapes human-computer interaction, migrating from purely functional systems to human-centric, empathetic environments represents a significant paradigm shift. Affective computing bridges this gap by enabling machines to recognize and adapt to human emotional states through computational modeling. This study investigates the detection of sequential patterns in physiological transitions during emotional episodes using the ARMADA algorithm, addressing the need for robust temporal analysis in smart, responsive ecosystems. By analyzing datasets with synchronized cardiovascular, electrodermal, and skin temperature features, the study validated ARMADA’s effectiveness for temporal pattern discovery. The generated association rules were found to be statistically significant via permutation tests, with over 50 rules meeting rigorous cross-dataset validation criteria. Quantitative analysis revealed significant findings regarding the predominance of moderate affective states, HRV synchronicity, and psychological lag in transition logic, offering actionable insights for the next generation of emotionally intelligent digital platforms.

ABSTRACT. Domain models have a myriad of uses, including analysis, requirement specification, aiding decision-making processes, facilitating shared understanding between stakeholders, and other applications. However, existing approaches and methodological guidelines for building such models are often either non-existent, isolated, ad hoc, incomplete, or only address specific parts of the domain modeling process rather than the whole endeavour. All of this has raised concerns about the validity, quality, and utility of domain models produced by such approaches, and the maturity of the domain modeling field as a whole. This study helps to address these issues by proposing a novel hybrid theory- and data-driven domain modeling framework that integrates taxonomy construction and computer-aided text analytics to develop practitioner-oriented classification dictionaries. The proposed methodological framework extends and improves upon seven existing guidelines from multiple fields by unifying them into a single, holistic, systematic, and generalizable approach. The proposed framework achieved 13.5/16 in a quality evaluation and was successfully applied to build a large-scale classification dictionary for the video game development domain. These results demonstrate the capability of the new approach to construct high-quality dual-purpose ontological and analytical models in complex domains.

ABSTRACT. Property-graph schemas often contain descriptive properties that recur across heterogeneous nodes and edges. Some are ordinary local attributes, while others behave as reusable metadata that may be externalized as shared structures. However, property-graph schema design lacks an explicit 5GNF-oriented method for deciding when a repeated property should become a reusable metadata candidate.

This paper proposes a rule-based design method for identifying such candidates before trait-node externalization. The method evaluates five criteria: cross-element occurrence, conceptual independence, lossless externalization, reuse potential, and governance relevance. Based on these criteria, properties are classified as trait candidates, embedded properties, or borderline cases.

The method is illustrated using a library-domain example and evaluated across two schema contexts. The validation combines eight human and four LLM-based responses and examines classification distributions, agreement, and comparison with a recurrence-based baseline. Results indicate that repetition alone is insufficient for externalization. Clear embedded properties and reusable descriptors achieve stronger agreement, whereas semantically ambiguous properties remain context-dependent. The method therefore provides an explainable basis for identifying reusable metadata candidates before 5GNF-oriented normalization.

ABSTRACT. Digital Transformation is driving significant changes in the port sector. Smart ports emerge as a response to these needs, integrating information technologies to optimise operations and minimise environmental impacts. Consequently, ports must deploy business process management mechanisms to coordinate and optimise workflows, enhance decision-making, and address the needs of the distinct stakeholders. We conducted a Design Science Research (DSR) project to develop, demonstrate, and evaluate a BPMN extension for smart and green logistics operations in ports. We collaborated with a port management entity to implement an innovative road entry terminal. Our extension incorporates elements that model (1) privacy, (2) energy efficiency, (3) risks, (4) data operations, and (5) domain-specific operations. The results are relevant to port business process modelling, as the developed extension aligns with the sector’s operational and logistical needs.

ABSTRACT. The growing importance of cybersecurity, particularly in network environments, necessitates advanced incident detection systems. Machine learning-based classification algorithms are crucial for identifying security breaches within network traffic. However, datasets representing this problem are typically highly imbalanced, which hinders the performance of standard classifiers. To address this issue, various oversampling techniques are commonly employed to increase the representation of the minority class, combined with an undersampling algorithm designed to refine the dataset by selecting the most informative minority class representations. This paper proposes extending this hybrid approach with a Transformer-based model. The effectiveness of the proposed approach is validated through computational experiments, demonstrating its potential for improving incident detection performance in imbalanced security scenarios.

ABSTRACT. We propose a stance classification framework based on hierarchically organized Differentiating Properties (DPs) -- linguistic features capturing opposing semantic attributes that define the poles of discourse polarization. DPs are automatically extracted from corpora via a multi-stage LLM pipeline and paired into oppositional variants, forming a bipolar semantic space. A linear Support Vector Machine (SVM) classifier operating on grouped DP embeddings performs stance classification by projecting target texts onto the polarization axis defined by opposing DP poles. Spearman rank correlation (ρ) and cosine similarity between pole embeddings define semantic convergence, serving as geometric diagnostic tools for assessing DP quality prior to classification. Experiments on three polarized corpora across two embedding models show that both metrics reliably predict quality of classification, as higher values correspond to higher F1 scores, while replacing high-ρ convergent pairs with low-ρ divergent pairs leads to a systematic performance degradation of up to 17%.

ABSTRACT. This paper does not aim to present a clinically ready Parkinson’s disease classifier. Instead, it rigorously examines what actually works in an extreme low-sample setting for bilateral, smoothed sEMG signals, using a dataset of only 9 subjects (5 Parkinson’s disease, 4 healthy controls; 18 full recordings, 122 segments). We audit classical subject-level baselines and multiple state-graph families under leakage-safe protocols, with seed-stability analysis, segment augmentation, paired tests, and a permutation sanity check.

A key finding is that isolated strong graph-model results can be misleading: some frozen graph configurations achieved ROC–AUC of 1.0, yet did not remain stable across seed reruns, and the best-performing graph branch changed across audited subsets. After fixing an evaluation bug, a naive subject-level baseline reached only 0.55 ROC–AUC, whereas a leakage-safe nested sparse baseline (outer and inner LOSO, fold-wise feature selection, logistic regression) achieved 0.778 accuracy, 0.775 balanced accuracy, 0.800 F1, 0.550 MCC, and 0.900 ROC–AUC, using 5.89 features per fold on average. The contribution is methodological: in ultra-low-N settings, evaluation rigor and stability analysis determine what counts as a valid result.

ABSTRACT. The growing availability of data distributed across multiple independent sources creates new challenges for classification tasks. Such datasets often differ not only in their representations but also in class proportions, which complicates the construction of a consistent global model. This work addresses these issues by extending the authors’ previously proposed distributed classification framework that combines conflict analysis, coalition formation, and rule induction. The novelty is achieved by introducing a class balancing stage performed locally on each dataset prior to the learning process. Six data-level balancing techniques are evaluated. Decision rules are generated using four rough set-based algorithms, and final decisions are determined by three strategies. Experiments are conducted on two datasets: Car Evaluation and Balance Scale. The proposed approach outperforms a baseline for Car Evaluation, while for Balance Scale its effectiveness depends on data fragmentation. Additionally, coalition analysis indicates that higher data distribution leads to larger coalition sizes.

ABSTRACT. Large multimodal models recently achieved impressive performance on a wide range of vision-language tasks, but training such models from scratch is computationally expensive and hardware demanding, which motivates fine-tuning and model‑reuse strategies. In this work, we present a practical example of reusing a pretrained image-text-to-text model (LLaVA) for predicting human gaze fixation heatmaps. Instead of instruction tuning with prompt and response pairs, we attach a gaze prediction model. In the presented method, we operate directly on the hidden representations of a frozen multimodal backbone that runs in a quantized configuration to reduce GPU memory and enable training on consumer-grade hardware. We also demonstrate that multimodal features can be effectively repurposed to map internal states to human gaze fixation from the LLaVA model, which was originally developed for text generation, after a relatively brief training period.

ABSTRACT. Effective electricity management on the consumer side is a key element in optimizing power system operation, particularly given the increasing share of renewable energy sources. Non-intrusive load monitoring (NILM) plays a crucial role by enabling detailed insight into energy consumption patterns. The obtained results enable load profile optimization by eliminating unnecessary consumption and shifting demand to off-peak periods and periods of high renewable generation. This paper presents a NILM system designed for real-world conditions characterized by the simultaneous operation of multiple devices and the presence of signal disturbances. The proposed approach is based on the synergistic use of steady-state and transient features, which serve as inputs to a multimodal deep neural network. The proposed method achieved an accuracy of 93.7%, confirming the effectiveness of multimodal feature fusion.

ABSTRACT. Public organisations increasingly adopt artificial intelligence (AI) to improve efficiency and data-driven decision-making, yet little is known about what AI does, in practice, to decision-making authority, accountability, and public value. Based on 14 semi-structured interviews with experts from national and county government organisations in Kenya, this study examines the tension between AI as a decision-support tool and AI as a mechanism of systemic control. Our results indicate that, alongside efficiency gains, AI shifts how public organisations operate, away from how they are intended to operate. Most importantly, we identify an emerging 'technocratic drift', in which decision-making authority and accountability shift from human-centred deliberation to technical optimisation and expert-driven systems. This prioritises measurable performance over accountability and citizen engagement, narrowing public value. We argue that AI adoption is far from a technical upgrade: it is a transformative shift that risks depoliticising governance and sidelining democratic oversight in favour of algorithm-driven control.

ABSTRACT. Today’s teenagers, who are part of Generation Z and Generation Alpha, make up “Generation Glass.” For this generation, the digital and traditional worlds are one and the same. They will soon become the recipients of public administration services. However, the adaptation of public services to the expectations of young citizens is proceeding very slowly. Thus, this article examines the potential of using low-code and no-code tools to develop public services. Additionally, this study identified areas where this technology can be applied in an urban context and the most significant benefits and limitations of its implementation. The study employed research methods, including a literature review to identify a research gap, as well as individual interviews with city representatives. This study may contribute to the accelerated implementation of new services for residents and support the city management process.

ABSTRACT. This paper presents a decision-support system for document routing in public administration, developed and deployed at the Patient Ombudsman Office in Warsaw, a central government authority in Poland. The proposed solution employs a convolutional neural network extended with a novel decision-making mechanism inspired by glial cells. In biological neural systems, glial cells perform regulatory functions by modulating neuronal activity and influencing the structure of neural connections. An analogous mechanism is incorporated into the model through dynamic control of feature map activity, enabling adaptive modification of a pre-trained network without full retraining. The main contribution of this work is a bio-inspired decision mechanism that allows selective regulation of feature representations in convolutional neural networks. The proposed approach facilitates the identification and elimination of low-informative feature maps, leading to structural optimization of the model. The system automates document routing by analysing document content and supports the processing of several thousand cases per day, significantly enhancing operational efficiency. In addition, the solution reduces document routing time and enables adaptive adjustment of the model to organisational changes. The system has been implemented as a component of an Electronic Document Management (EZD) system and operates in a production environment.

ABSTRACT. Supply chains face growing uncertainty in turbulent environments characterized by pandemics, geopolitical conflicts, and energy crises. This paper investigates how these disruptions shape uncertainty perceptions among supply chain actors and what this implies for IS operations and management. A quantitative survey of 200 medium and large companies across four sectors in Poland assessed 64 uncertainty factors using a five-point Likert scale. Respondents were geographically distributed across all 16 Polish voivodeships, enabling consideration of regional conditions. Results indicate that demand-related factors were rated most impactful, while wars, CO2 prices, and weather were systematically underestimated—yet proved critical to supply chain disruptions post-survey. Sectoral differences were statistically significant for only select factors, suggesting broadly shared uncertainty perceptions across industries. These findings highlight critical gaps in IS operations management, particularly the need for agile, resilience-oriented decision-support systems capable of responding to low-probability, high-impact events with spatially differentiated characteristics.

ABSTRACT. Selecting personal banking products requires evaluating multiple criteria under changing market conditions. Traditional multi-criteria decision analysis (MCDA) approaches are typically static and fail to account for temporal dynamics. This paper proposes a time-aware decision support system that extends the SPOTIS method by integrating longitudinal data into a unified evaluation framework. The system enables simultaneous modeling of criteria importance and temporal preferences, such as recency versus stability. An empirical study of Polish banks demonstrates that the proposed approach provides more informative and context-sensitive rankings compared to conventional methods. The results highlight the value of incorporating temporal information into decision support systems for dynamic environments.

ABSTRACT. The aim of the article is to assess the level of competencies of managers of digitized production enterprises operating in the Polish agricultural machinery sector, with particular emphasis on the production of components and subassemblies of agricultural vehicles. In light of the ongoing digital transformation of industry, the need was indicated for identifying the competencies necessary for effective management under Industry 4.0 conditions, as well as determining directions for adaptive actions for enterprises and academic institutions training future managers. Achieving the research objective required identification of key competency areas based on analysis of relevant literature, development of a research model in the form of a competency assessment sheet, and evaluation of the desired and actual levels of competencies among managers. The research results form the basis for identifying competency gaps and formulating recommendations regarding directions for the development of managerial competencies in the context of the digital transformation of industry.

ABSTRACT. The rapid integration of Large Language Models (LLMs) is transforming how developers search for information and solve technical problems. While these tools offer powerful support, they may also encourage cognitive offloading, potentially affecting independent problem-solving. This study examines whether LLM adoption relates to changes in troubleshooting traces in Stack Overflow questions. Drawing on cognitive offloading, transactive memory theory, and SRL, we analyze help-seeking behavior before (2019–2021) and after (2023–2025) widespread LLM adoption. Using machine learning–assisted annotation, we develop a Learning Trace Index (LTI) to capture observable problem-solving processes. Results show a sharp decline in platform activity after 2023, but only modest differences in troubleshooting traces across periods and experience levels. These findings offer empirical insight into cognitive offloading and introduce a scalable measure for studying learning behaviors in developer communities.

ABSTRACT. Healthcare organisations face a critical gap in AI evaluation: accuracy benchmarks do not indicate whether systems will function reliably within the constraints of real clinical environments. This study addresses that gap with two contributions. First, it introduces the enrollability framework, four Information Systems-level criteria, grounded in Actor-Network Theory, that define when a Natural Language Processing (NLP) method can be stabilised as a clinical infrastructure. Second, it examines how Large Language Models (LLMs) align with these requirements in a non-English clinical NLP setting. While rule-based systems encode stable inscriptions compatible with network expectations, LLMs rely on next-token prediction that mimics but does not ensure such stability. We empirically compare rule-based and LLM-based approaches (Llama-3-8B-it, Gemma-7b-it) on 1,679 Polish paediatric epicrises. Findings reveal that accuracy and enrollability yield divergent evaluations, and suggest that common LLM failures (e.g., hallucinations, inter-physician variability, and notation mismatch) may be associated with properties of their generative mechanism, beyond capability limits.

ABSTRACT. Integrating sustainability into software engineering education is essential for preparing future professionals for digital transformation. This paper presents a study with a pre-post design that evaluates the impact of embedding sustainability within a software quality assurance assignment in an undergraduate course. Results show an improvement in students' understanding, shifting their perspective from a technical to a more holistic, multi-dimensional view of sustainability. The intervention increased their confidence in interpreting sustainability metrics and enhanced practical skills, although challenges related to the adoption and use of sustainability metrics persist.

ABSTRACT. The study investigates the determinants of individual acceptance of agentic AI by integrating traditional technology acceptance constructs with agency-related characteristics of AI agents. While classical models such as UTAUT have been validated for traditional IT systems, emerging agentic AI, due to its autonomous performance and decision-making, requires a different theoretical lens. We develop and test an extended acceptance model incorporating perceived competence, decision autonomy, and performance autonomy alongside traditional determinants. A quantitative survey of AI users was conducted, and the collected data were analyzed using PLS‑SEM. The results demonstrate that all three agentic characteristics significantly shape performance expectancy, which, together with facilitating conditions and social influence, predicts intention to use AI agents. The findings highlight the need to reconceptualize acceptance frameworks for increasingly autonomous AI systems.

ABSTRACT. The pharmaceutical industry is facing a significant transformation with the emergence of Industry 5.0, shifting its focus towards human-centeredness and sustainability. The right leadership approach is crucial for the successful implementation of this transformation. This exploratory study examines the approaches needed in the context of Leadership 5.0 adoption and its role in advancing human-machine collaboration and sustainable innovation in the pharmaceutical industry. By presenting a network analysis of the scientific literature, it maps the position of the Leadership 5.0 concept in the literature, its relationships with other concepts, and identifies research gaps. The analysis, based on a systematic review of 40 articles from Scopus and Web of Science, shows that research on Leadership 5.0 in the pharmaceutical industry is still in its early stages. This exploratory study lays the groundwork for future empirical research and may support managers and academics in understanding Leadership 5.0 adoption.

ABSTRACT. The aim of this article is to identify patterns of co-occurrence in assessments regarding implementation determinants, benefits, barriers, post-implementation changes, and organizational outcomes associated with the use of Cloud Business Intelligence (CBI), as well as to identify enterprise segments with similar perceptions of these phenomena. The study was based on data obtained from 400 enterprises using cloud-based business analytics solutions. The analysis employed an exploratory approach, combining Multivariate Correspondence Analysis (MCA) with Ward's hierarchical cluster analysis. The results indicate that perceptions of CBI implementation have a multidimensional, non-random structure, and respondents' responses form distinct patterns that differentiate organizations in terms of the observability of benefits, the level of perceived implementation effectiveness, and the exposure to barriers. Based on this, five enterprise segments were identified, differing not only in their assessment of implementation outcomes but also in their organizational, technological, and decision-making profiles.

ABSTRACT. Governance Risk and Compliance (GRC) activities are conducted by employees with high educated staff. The assets to be audited undergo control activities that are carried out by focusing on the areas to be audited within the audition space. In order to reach the analysis results of the AS-IS and TO-BE (new) process models with simulation, the effort estimates of these processes were obtained with the fuzzy AHP. Accordingly, it has been determined that information technology (IT) solutions with software developments to be used in audit activities can increase the completion time efficiency by 51.7%. In addition, it has been obtained that the workload of human resources decreased from 61% to 49%. The average risk occurrence probability of 5.85% has been reduced to 3.90%. The monetary loss of risks could be reduced 30.54%. In the creation of the new Audit Execution Process, the simulation of the AS-IS process is an indicator of the originality of this study.

ABSTRACT. This paper presents a digital twin architecture that mirrors how real admission processes work and shows where AI can be safely introduced. The system reconstructs workflows from historical data and uses AI agents to analyze documents, video, and structural data. Instead of relying on black box LLM based predictions, the system uses simple decision trees at each step of the process. These models make decisions explicit and easy to audit. A state based orchestration layer connects AI outputs to clearly defined process stages and supports a transition from notebook based prototypes to enterprise systems. The key contribution of this work is a practical development approach for AI systems. Instead of building monolithic solutions, the business logic is decomposed into small AI agents implemented as executable notebooks that communicate through cloud storage. This prototype can then be automatically transformed into enterprise ready services using AI programming agents. The approach is evaluated on real applications from a university admission process. Results show that the digital twin closely reproduces historical decisions while remaining interpretable. This work addresses the challenge of deploying AI in high stakes systems where both performance and explainability are required.

ABSTRACT. Decision support systems embedded in modern information systems often rely on multi-criteria decision analysis (MCDA) to evaluate alternatives; however, selecting an appropriate MCDA method and managing imprecision in stakeholder assessments remain persistent challenges. The Iterative Compromise Ranking Analysis (ICRA) addresses the former by compromising rankings from multiple MCDA methods, but its original formulation assumes crisp inputs. This paper extends ICRA to operate on triangular fuzzy numbers (TFNs), enabling the processing of imprecise expert evaluations under uncertainty. The proposed Fuzzy ICRA applies alpha-cut arithmetic to propagate uncertainty through the MCDA methods at each iteration, preserving the informational content of fuzzy inputs throughout the compromise process. A simulation study with three experiments examines the method using TOPSIS and VIKOR. A step-by-step example demonstrates convergence and uncertainty evolution; a propagation analysis across six spread levels shows that ICRA reduces input uncertainty over iterations; and a robustness analysis evaluates sensitivity to expert-assessment noise. Results indicate reliable convergence and a consistent uncertainty-reduction effect, with TFN spreads decreasing monotonically after the second iteration.

ABSTRACT. We propose a novel preference-based method for generating counterfactual explanations in AI-powered information systems that explicitly accounts for individual user preferences. To the best of our knowledge, our method is the first preference-based counterfactual generator in this area. We have conducted experiments on three popular datasets (Communities and Crime, German Credit, and Default of Credit Card Clients) with continuous attributes and binary labels. For those experiments, our method is feasible and capable of producing more relevant counterfactuals than other methods. In the given settings, it is also superior in terms of computational efficiency. In all cases we examined, our method outperformed other techniques in terms of preference alignment. These findings demonstrate the potential of preference-aware counterfactual explanation generation to enhance the adaptability and relevance of intelligent information systems, especially in the context of the sustainable integration of AI and automation.

ABSTRACT. An accurate representation of decision-maker preferences is a critical requirement for modern information systems and decision support systems, particularly in environments characterized by uncertainty. To address this challenge, this study proposes an extension of the Fuzzy Ranking Comparison (RANCOM) method by introducing alternative relationship functions and a controllable uncertainty parameter $\gamma$, enabling more flexible modeling of subjective judgments. A simulation-based analysis was conducted to evaluate the impact of different relationship functions on the accuracy of representing fuzzy criteria weights using multiple performance metrics. The results show that the logistic function consistently achieves the lowest TFN distance and the highest TFN similarity across most configurations, indicating the most accurate reconstruction of the reference preference structure. Additionally, the ratio-based and exponential functions outperform the standard three-value scheme, particularly for smaller problem sizes. These improvements enhance the ability of decision support systems to operate under uncertainty, leading to better alignment with user preferences and more reliable decision outcomes. The findings highlight the potential of the extended Fuzzy RANCOM method as an effective component of advanced information systems.

ABSTRACT. While Coronary Angiography is essential for AI diagnostics, ensuring data privacy within healthcare Information Systems remains challenging, as conventional anonymization fails to eliminate hardware-specific forensic signatures. Addressing the problem of how to balance strict privacy with downstream diagnostic utility, we propose a novel obfuscation pipeline. Instead of traditional image editing, we synthesize angiograms derived solely from binary vessel masks. This removes forensic artifacts while retaining the vascular structures vital for clinical diagnosis. We evaluate this against an Attribute Inference Attack identifying institutional data sources, comparing Pix2Pix and Diffusion architectures against classic augmentations. The proposed methods drop attack success from $F1=0.8$ to $F1\approx 0.3$ while maintaining binary segmentation $F1\approx 0.64$ (Pix2pix) and $\approx 0.79$ (Diffusion). Results demonstrate that mask-conditioned reconstruction significantly reduces the risk of source attribution, offering a means of anonymized medical data sharing across inter-organisational IS without a significant diagnostic utility trade-off.

ABSTRACT. Systematic literature reviews (SLRs) commonly use keyword co-occurrence analysis for thematic mapping, but standard tools (e.g., VOSviewer, Bibliometrix) are limited to symmetric pairwise relationships. The aim of this study is to formalize author keyword association rule extracting (ARM) as a repeatable, complementary method for thematic mapping in SLR and to evaluate its analytical value relative to standard co-occurrence methods. We apply the Apriori algorithm to author keywords from Scopus and demonstrate it on two datasets: a medium one (597 publications on CSR and consumer behavior) and a large one (14,302 publications on gamification and marketing). Comparing ARM results with VOSviewer co-occurrence maps, we find ARM reveals directional dependencies and multi-item themes that symmetric analyses miss. We propose a three-step framework to incorporate ARM into the PRISMA synthesis stage and discuss using ARM-derived themes for embedding-based screening in EmbedSLR.