Purpose Mobile robot path planning is a fundamental task in autonomous navigation. Conventional ant colony optimization (ACO) and some of its variants may exhibit weak search guidance, slow convergence, and redundant initial paths in complex grid environments. This study proposes a hybrid heuristic ant colony optimization algorithm, termed HHACO, for improving path quality and search efficiency in known static grid environments. Design/methodology/approach HHACO integrates weighted Euclidean distance, Manhattan distance and historical global-best path proximity into heuristic information. A dynamic balance factor and selection threshold regulate node-selection behavior across iterations. A dual-path pheromone update jointly reinforces the current iteration-best and historical global-best paths under pheromone bounds. After a complete path is generated, local line-of-sight optimization removes redundant nodes subject to collision checking. Findings Parameter, ablation and comparative experiments on several static grid maps show that HHACO obtains shorter feasible paths and favorable performance in convergence iterations, turning points, and result stability in the tested scenarios. Comparisons with recent representative multi-strategy ACO methods further indicate faster convergence or improved path quality in the corresponding environments. Originality/value This study improves conventional ACO from the perspectives of heuristic information, state transition, pheromone updating and path post-processing for path planning in known static grid environments. The proposed method combines hybrid geometric guidance, historical high-quality path guidance, dynamic search control, dual-path pheromone updating and local line-of-sight optimization, providing an improvement approach for enhancing path quality and search efficiency when a certain amount of additional computational cost is acceptable.
Purpose Online learning environments provide rich digital traces of learner activity, yet they offer limited access to the non-verbal affective cues that instructors naturally observe in face-to-face classrooms. This study investigates facial expression analysis as an affective sensing component for emotion-aware online learning. Design/methodology/approach Facial data were collected from 26 students during online learning sessions, and self-reported emotion labels were mapped into three affective categories: positive, neutral and negative. The visual stream was transformed into standardized face-centered representations and evaluated using a lightweight CNN implementation together with representative pretrained CNN architectures. Findings The results show that facial expressions are perceived by participants as meaningful non-verbal cues in online learning. In the classification experiments, VGG19 achieved the highest accuracy (0.79), while the lightweight CNN achieved a comparable accuracy (0.78) with the lowest loss value. Originality/value These findings suggest that facial-expression-based affective cues can be extracted from online learning data and may complement conventional learning analytics in future emotion-aware educational systems. Highlights
Purpose Integrating explainable artificial intelligence (XAI) for water quality assessment (WQA) is necessary to protect human health, ensure the availability of clean water and sustainable conservation of the environment. Black-box machine learning (ML) models perform well but lack cognitive insight, limiting their use in decision-support systems. This research addresses the need by putting forward a concise, explainable technique for binary classification of water quality. The research presents a comprehensive autonomous decision-making framework for assessing water quality, incorporating XAI, feature selection and fuzzy IF-THEN reasoning. Design/methodology/approach To ensure the integrity of the statistical analysis, missing data has been addressed through the implementation of multiple imputation by chained equations (MICE). Effective use of highly performing classifier facilitated the predictions. Shapley additive explanations (SHAP) were adopted to identify and normalize significant characteristics. To boost interpretation, fuzzy linguistic terms are generated using arcsinh-based quartile partitioning, facilitating the formation of SHAP-driven fuzzy IF–THEN rules. The validity of the rules has been monitored by activation strength analysis to confirm the consistency between fuzzy inference and predictions of the models. Findings Outperforming all other models, the Random Forest algorithm scored highest accuracy of almost 78%. Through the integration of SHAP computation, the most influential criteria of water quality were determined. The fuzzy modeling process was made simpler by rule creation based on the most essential variables with time complexity reduction of 46.9%. The established rules offered clear conclusions about the evaluation of water potability and successfully translate complicated ML results into indicators that humans can grasp. Selection of model with high accuracy, activation strength analysis of rules, inter-fold consistency in SHAP ranks demonstrates that the suggested framework attains both predictive and interpretative stability. The proposed XAI–SHAP black-box fuzzy model facilitates the development of autonomous decision-making framework for public health, water quality and environmental sustainability. Research limitations/implications Despite the scientific merits of soundness and understanding of the described SHAP-fuzzy architecture, numerous limitations have to be admitted. The RF model had experienced medium predictive accuracy with an accuracy of about 78% throughout cross-validation. The standard deviation is relatively small indicating the model is stable and is always generalizing correctly. On the other hand, it may be possible to enhance the classification effectiveness with better advanced data pre-processing, dealing overlapping of criteria and class imbalance and use of integrated models, which have a higher predictive strength. Practical implications The framework can support water management authorities in making faster, more transparent decisions about water quality. By combining explainable AI with fuzzy reasoning, it helps non-experts understand why certain assessments are made, improving trust and accountability. It can be applied to real-time monitoring systems to detect contamination risks early and prioritize interventions. The approach also enables efficient resource allocation by focusing on the most influential parameters. Social implications The framework can improve public health by enabling earlier detection of water contamination and more reliable quality assessments. Its transparency helps build trust among communities, regulators and stakeholders by clearly explaining decisions. Better water management can support equitable access to safe water, particularly in vulnerable regions. However, disparities in data availability and technical infrastructure may widen gaps between well-resourced and underserved areas. Originality/value In contrast to traditional methods that utilize SHAP exclusively as an interpretation instrument, the suggested framework developed a trustworthy methodology by integrating a transparent connection between ML explanations and linguistic fuzzy reasoning. The proposed XAI–SHAP fuzzy model facilitates the development of decision-making framework for public health, water quality and environmental sustainability. It uniquely combines SHAP-based feature normalization, arcsinh-based quartile partitioning, understandable IF-THEN fuzzy rules and activation strength analysis. This guarantees both interpretability and stability, offering a clear and elucidative method for binary classification of water quality.
Purpose Click-through rate (CTR) prediction is a core challenge in digital advertising, as forecasting whether a user will click on an advertisement directly determines ad placement decisions and revenue outcomes. Standard machine learning and deep learning approaches achieve reasonable predictive accuracy but are largely opaque, making it difficult to determine which features drive predictions and why. Deep Neural Networks (DNN), Deep Factorization Machines (DeepFM), and Deep Cross Networks (DCN) each capture different aspects of feature interaction, yet none alone addresses the full complexity of large-scale sparse data. Design/methodology/approach This paper proposes a stacking ensemble learning model for click-through rate (SELM-CTR) prediction, a stacking ensemble that combines DNN, DeepFM, and a Gated Deep Cross Network (GDCN) as base models, with XGBoost serving as the meta-model. Rather than treating these architectures as interchangeable alternatives, we exploit their complementary strengths: the DNN learns nonlinear representations, DeepFM captures low- and high-order interactions through factorization, and GDCN applies gated cross-layer interactions. The meta-model is trained on out-of-fold predictions from the base models, allowing it to learn which architecture to trust for different input patterns. Findings Evaluation using the publicly available AVAZU dataset indicates that the proposed method achieves an accuracy of 89%, an Area Under the Curve (AUC) of 94%, and a log loss of 0.25. These results represent a measurable improvement over existing baseline approaches. Furthermore, the application of SHAP clarifies how specific features influence the model's predictions, providing practical insights for real-world decision-making. Originality/value The primary contribution of this work is the integration of a stacking ensemble architecture (using DNN, DeepFM, GDCN, and XGBoost) with SHAP-based feature analysis. This addresses the common “black-box” limitations of deep learning in advertising, ensuring both high predictive performance and greater transparency regarding feature contributions.
Purpose Nowadays, fuzzy modeling has been technically considered as a meaningful way to address uncertainty in a wide range of applications. Hence, it can be used to target a classic uncertainty related to the best choice of the trust region (TR) ratio, recognized as an influential factor that highly affects the computational performance of the TR algorithms. Design/methodology/approach Determined according to the quality of the approximate local model of the given unconstrained minimization problem, researchers have justified the pros and cons of their choices for the TR ratio. By the way, an optimal choice for the TR ratio is still a matter of further study. Here, we plan to simultaneously benefit from the advantages of several effective choices of the TR ratio by first employing them to form a fuzzy number and then generating several quantifiable results as hybrid formulas for the TR ratio by applying the fuzzy ranking functions. Moreover, we suggest a local diagonal approximation for the Hessian to be used in the TR subproblem and make it possible to handle high-dimensional optimization models. Findings We analytically assess the effect of our modifications on the global convergence of an adaptive nonmonotone TR algorithm. We also conduct some computational experiments to provide further justification for our theoretical improvements. The results emphasize that the fuzzy concepts can helpfully make progress in the performance of the TR algorithms. Originality/value We propose a two-part method to enhance TR algorithms. First, we combine multiple update strategies for the TR radius using fuzzy logic to create a robust hybrid rule. Second, to ensure scalability, we approximate the Hessian matrix with a simple diagonal form, making the approach efficient for high-dimensional problems.
Purpose The purpose of this study is to investigate the potential of integrating metaverse technologies into sharing economy systems to enhance sustainability, optimize resource utilization and improve operational efficiency, particularly in the transportation sector. In this context, the study specifically addresses the challenges of modeling complex and interdependent decision information under uncertainty. To achieve this, an advanced multi-criteria decision-making framework under a complex intuitionistic fuzzy (CIF) environment is developed for systematic evaluation. Design/methodology/approach This study proposes two novel aggregation operators, namely the complex intuitionistic fuzzy reducible weighted Maclaurin symmetric mean (CIFRWMCSM) and the complex intuitionistic fuzzy reducible weighted dual Maclaurin symmetric mean (CIFRWDMCSM), to address the shortcomings of existing CIF-weighted Maclaurin symmetric mean (MCSM) operators, such as the lack of idempotency and reducibility. The mathematical formulations, fundamental properties and special cases of the proposed operators are rigorously investigated. Furthermore, the weighted aggregated sum product assessment (WASPAS) method is extended by incorporating the proposed operators to handle CIF-based decision-making problems. The applicability and effectiveness of the developed framework are demonstrated through a real-world case study on metaverse-enabled sharing economy solutions in transportation. Findings The results reveal that the proposed CIFRWMCSM- and CIFRWDMCSM-based WASPAS framework provides stable, consistent and reliable ranking outcomes while effectively capturing interrelationships among decision criteria. The proposed approach also ensures improved aggregation consistency due to the satisfaction of reducibility and idempotency properties. Sensitivity analysis and comparative studies confirm that the proposed approach exhibits superior robustness and performance compared with the existing method. Practical implications The proposed framework offers decision-makers a practical and reliable tool for evaluating metaverse-driven sharing economy initiatives, supporting informed decision-making aimed at sustainability improvement, efficiency enhancement and safer transportation systems. It is particularly useful in scenarios involving complex, uncertain and multi-dimensional decision environments. Originality/value This study introduces reducible and idempotent CIF-based MCSM aggregation operators for the first time and integrates them into an extended WASPAS framework. By resolving key limitations of existing MCSM operators and enabling the modeling of higher-order attribute interactions, the proposed approach advances the theoretical development of CIF aggregation operators and provides a novel, robust decision-support methodology for assessing metaverse-enabled sharing economy applications.
PurposeThis study provides a comprehensive review of Türksen's Fuzzy Functions (FF) approach, examining its methodological evolution and research landscape over nearly 2 decades to identify theoretical advances, application patterns and future research directions. Design/methodology/approachA comprehensive systematic literature review with bibliometric and social network analysis is conducted. The methodological review synthesizes 76 publications. The developments are categorized into seven evolutionary pathways: early developments, clustering enhancements, evolutionary computation integration, higher-order uncertainty modeling, recurrent architectures, robustness considerations and ensemble learning. Bibliometric analysis examines productivity patterns, citation structures, author collaboration networks, institutional distributions and thematic organization. FindingsThe FF approach has evolved from basic FF-LSE models to sophisticated architectures incorporating advanced clustering algorithms, evolutionary optimization, type-2/intuitionistic/picture fuzzy sets, recurrent structures and robust estimation. Bibliometric analysis reveals cyclical productivity with peaks in 2020 and 2022, strong geographic concentration in Turkish institutions, fragmented collaboration networks with minimal cross-institutional bridges and application focus on financial forecasting. Critical gaps include incomplete theoretical formalization of higher-order variants, limited international diffusion and insufficient attention to model interpretability. Originality/valueThis study presents the most comprehensive bibliometric analysis of FF research, systematic categorization of methodological innovations, quantitative mapping of application domains and integrated identification of theoretical gaps and methodological challenges, establishing a structured research agenda for future efforts.
Purpose In response to the inherent ambiguity of requirements and the suboptimal efficiency in decision-making processes related to age-friendly design aspects of urban bus stops, a comprehensive framework driven by an integrated Kano-TOPSIS approach was proposed. Design/methodology/approach Utilizing data research and user interviews, a pool of user demands was constructed, leading to the identification of 19 characteristics of waiting requirements for the elderly population. Subsequently, the Kano model was applied for classification and selection, establishing a hierarchical structure of age-friendly requirements for bus stops, which served as the foundation for proposing two age-friendly bus stop design proposals. With the integration of the TOPSIS multi-criteria decision-making model, the design proposals and samples were subjected to decision calculations and evaluations to determine their priority ranking. Findings The results indicated that the two age-friendly bus stop designs developed based on high-weight design factors selected by the Kano-TOPSIS model outperformed existing relevant designs. The Kano-TOPSIS model accurately identifies hierarchical elderly user needs and transforms them into a quantified ranking of design elements, providing a scientific basis for decision-making in the age-friendly design of urban bus facilities. Originality/value By analyzing the behavior patterns and differences in demands of the elderly population, this research drives the refined categorization of application scenarios for public products, paving the way for future innovations in age-friendly development of public products.
Purpose This study aims to propose a complete data-driven framework for the Transit Network Design Problem (TNDP) using large-scale taxi GPS trajectories to infer travel patterns and optimize transit routes. Design/methodology/approach The proposed framework consists of two stages. First, a convolutional weighted multi-objective stop selection (CWMSS) algorithm is introduced to identify high-demand stop areas by integrating passenger demand, coverage strength, and walking distance using a distance exponential decay mechanism. Second, a Deep Q-Network (DQN) combined with a Graph Neural Network (GNN) is developed. This model uses a fine-grained neighbor-extension strategy to incrementally construct transit routes, optimizing a tunable multi-objective cost function that balances passenger experience and operational cost. Findings Evaluations using a decade-long New York City (NYC) taxi dataset show the CWMSS algorithm improves Pickup and Drop-off Records (PDRs) and Effective Coverage Area (ECA) by 9.39% and 21.07% against state-of-the-art stop selection methods. Furthermore, the DQN-GNN model outperforms two advanced heuristics and two state-of-the-art Reinforcement Learning (RL) baselines in passenger-operator balanced and passenger-oriented scenarios. Friedman and Wilcoxon signed-rank tests indicate that the observed improvements are statistically significant (p < 0.05). Originality/value This research addresses the limitations of applying RL to TNDP on large-scale real-world networks. It contributes a novel stop selection algorithm using convolutional operations and a DQN-GNN model. This approach not only resolves the curse of dimensionality inherent in large-scale urban environments but also successfully designs transit networks that simultaneously improve passenger experience and lower operational costs.
Purpose This study presents a new startup investment decision support model (SIDSM) that is specifically designed to mitigate the inherent challenges of class imbalance and uncertainty in startup investment decisions. Design/methodology/approach Based on financial and structural indicators sourced from the Crunchbase database, the proposed model incorporates a multi-stage methodology. First, a systematic feature selection process integrating SHAP, Boruta and Elbow methods is used to retain informative features. Subsequently, uncertainty estimates are calculated at the feature and observation levels using the Shannon-Entropy and DeepGini metric and included in XGBoost's learning process through a user-defined loss function and a label-distribution-aware margin (LDAM) integration. Cuckoo Search meta-heuristic algorithm is used to optimize the hyperparameters to ensure model robustness, and a class-based threshold optimization is used to optimize decision boundaries. Findings The experimental findings demonstrate that SIDSM outperforms the baseline models, achieving a macro F1-score of 89.47% and more stable minority class detection, thereby indicating its potential to support startup investment decisions in a reliable, transparent and evidence-driven manner under class imbalance. Originality/value This study proposes a novel, data-conscious and holistic approach to the methodological limitations of traditional classification approaches. By integrating feature explanatory power and observation-based uncertainties into the learning process, rather than relying solely on error-driven optimization, the model becomes sensitive to uncertainty, captures difficult-to-learn and minority-class patterns better and provides a highly robust and explainable framework for startup investment decision-making.
Purpose Self-Admitted Technical Debt (SATD) consists of source-code comments in which developers explicitly acknowledge suboptimal design or implementation decisions that require future improvement. These comments often convey emotional signals such as frustration, urgency, or concern, which may reflect the perceived severity and priority of technical debt. While sentiment analysis has been increasingly applied to SATD, little attention has been paid to the interpretability and reliability of sentiment predictions produced by modern deep learning models. This study aims to investigate how explainable artificial intelligence (XAI) techniques interpret SATD sentiment predictions and whether different model-agnostic explanation methods provide consistent or divergent explanations. Specifically, we examine the reliability, agreement, and limitations of popular post-hoc explainers when applied to BERT-based SATD sentiment classification. Design/methodology/approach We formulate SATD sentiment analysis as a binary classification task that distinguishes negative from non-negative comments and fine-tune a BERT model on a manually curated SATD sentiment dataset using ten-fold cross-validation. For all correctly predicted instances, we generate local token-level explanations using three model-agnostic XAI techniques: LIME, SHAP, and BreakDown. We quantitatively assess explanation behaviour and cross-method consistency using feature contribution distributions, top-k token overlap, semantic similarity based on BERT embeddings, and Spearman rank correlation. Findings The results reveal substantial divergence among the three explanation methods. LIME, SHAP, and BreakDown assign markedly different contribution magnitudes to influential tokens, exhibit near-zero overlap in top-k features, and frequently produce contradictory ranking orders and sentiment contribution directions. Statistical tests further confirm that these differences are systematic rather than random across both negative and non-negative sentiment categories. Originality/value Our findings demonstrate that model-agnostic explanation techniques cannot be used interchangeably for interpreting SATD sentiment predictions. Relying on a single explainer may lead to incomplete or misleading interpretations of developer intent. We therefore recommend multi-method triangulation and manual validation when explanation results are used to support technical debt prioritisation, code review, or maintenance decision-making. This study provides a comprehensive and reproducible empirical analysis of explanation reliability and divergence for SATD sentiment analysis, contributing a foundation for trustworthy and interpretable SATD analytics.
Purpose Digital platforms across the globe and specifically in India increasingly demand recommender systems that elevate user experience, empower producers and reassure platform custodians simultaneously. We therefore tackle data sparsity, cold-start fragility, non-linear preference formation and opaque decisioning without compromising creator exposure as well as systemic health. Design/methodology/approach Heterogeneous ensemble with inclusion of contrastive self-supervision for solving sparsity, transformer-based sequence modeling to solve behavioral evolution, and a TabNet deep tabular learner for heterogeneous features is coordinated by a context-aware meta-learner. Training optimizes three interconnected objectives, maintaining user fidelity, enhancing producer opportunity and ensuring overall system balance. Stakeholder telemetry is emitted each validation pass through an automated pipeline that acts as a deployment gate. Findings On MovieLens-100K, 1M and 20M, the ensemble achieved user-centric Root mean square error of 0.9473, 0.8942, and 0.8651, respectively, with normalized discounted cumulative gain consistently above 0.98. Producer-centric metrics climb from an exposure parity of 0.0059 and catalog reach of 0.0059 on MovieLens-100K to 0.0421 and 0.0385 on MovieLens-20M, with Hit Rate@10 undefined on the smallest split but measurable at 0.0245 and 0.0389 on the 1M and 20M datasets. System-centric evaluation maintains bias-free coverage above 0.996 and interaction diversity between 0.3666 and 0.4123, while stability is only reported for the 1M and 20M datasets where temporal windows are populated. Multi-stakeholder optimization yields combined scores of 0.4304, 0.5127 and 0.5938 across datasets, proving coordinated stakeholder benefit. Originality/value This study introduces the multi-stakeholder recommender system that is benchmarked on the complete MovieLens dataset, embedding producer and system telemetry inside the optimization loop. The use case of the solution creates by us lies in the domain of over-the-top (OTT), ed-tech and commerce ecosystems pursuing responsible artificial intelligence.
Purpose This paper addresses challenges in high-density higher education classrooms, including severe occlusion and concealed disciplinary behaviors. It aims to propose the WM-YOLO network and a vision-based quality evaluation system for precise micro-behavior detection and macro-quality assessment. Design/methodology/approach We propose an improved WM-YOLO network that integrates Weighted Convolution (wConv2d) and a Multi-scale Attention Network (MANet). wConv2d suppresses occlusion noise, while MANet improves small-target detection. To validate this architecture, the research utilizes a self-constructed High-Density Classroom Behavior dataset (avg. 51 students/frame). Furthermore, the study redefines the Effective Heads-up Rate (HuR, denoted R_eff) and Negative Engagement Index (I_neg) to analyze classroom state dynamics. Findings Experiments show that WM-YOLO achieves an mAP@50 of 92.3%, outperforming baselines by 1.4% with 27% lower computation. Empirical analysis reveals a “Negative Scissors Difference” (heads-down exceeding heads-up) during non-teaching periods, validating the model’s ability to distinguish classroom states. Practical implications The proposed evaluation model effectively supports higher education administration. By identifying abnormal fluctuations in I_neg and the “Scissors Difference” phenomenon, administrators can automatically filter out non-teaching durations, enabling automated attendance tracking and precise teaching quality warnings, thus promoting a shift from “experience-based management” to “data-driven governance”. Originality/value This paper innovatively applies wConv2d and MANet to high-density scenes. It establishes a quantitative link between micro-behaviors and macro-evaluation via the “Scissors Difference” principle, offering a new paradigm for smart education.
PurposeThis work presents an overarching data-driven evaluation of the viability of bioplastics as alternatives to petroleum-based shopping bags, considering multiple criteria associated with the sustainability agenda. Subsequently, it proposes a novel integration of stochastic multi-criteria acceptability analysis (SMAA) and combinative distance-based assessment (CODAS), hereby coined as SMAA-CODAS, to identify the most preferred bioplastic substitute.Design/methodology/approachThorough review of the domain literature with inputs from an expert team identified 12 criteria, comprehensively spanning environmental (i.e. composition of bio-based materials, biodegradability through natural processes, biodegradability through industrial processes, feedstock sustainability, greenhouse gas (GHG) emissions and water footprint), technical (i.e. fitness of domain application, bulk density, tensile strength and impact strength) and economic (i.e. market price and market demand) attributes. The dataset comprising the evaluation of 12 bioplastic alternatives was extracted from different sources in the literature, including journal articles, reports and policy documents, with data expressed as real, interval or fuzzy numbers to capture the uncertainty arising from inconsistencies in data across sources. In the proposed approach, each Monte Carlo simulation run constructs a real-valued decision matrix drawn from the imprecise dataset. Subsequently, CODAS implements the evaluation, while SMAA implements multiple runs and reports insightful statistics as outputs.FindingsAfter 10,000 runs, polyglycolic acid (PGA) emerged as the most robust alternative, ranking first 85.39% of the time (rank acceptability index) and achieving the highest holistic acceptability index, followed by polylactic acid (PLA) and polyhydroxyalkanoates (PHA), while polyethylene terephthalate made from bio-based sources (Bio-PET) ranks last. Also, the central weight vector shows that PGA, PLA and PHA were primarily contributed to by biodegradability through industrial and natural processes and by the composition of bio-based materials, except for PGA, which had the lowest GHG emissions. Comparative analysis with other SMAA variants yields a correlation coefficient of at least 0.92. Sensitivity analysis shows that the evaluation of bioplastics is highly influenced by the criteria weights, allowing policymakers to reflect their preferences across specific scenarios.Practical implicationsFindings from this work suggest that research and development of PGA must be a priority in investment decisions, including feedstock choice, production scalability and cost-optimization measures. Although PGA emerges as the most preferred choice, its current commercial deployment is constrained by relatively higher production costs and limited large-scale manufacturing infrastructure compared to conventional plastics and more established bioplastics such as PLA. Thus, when cost trade-offs are substantial, PLA may be considered a second-preferred alternative, as it exhibits properties closely comparable to PGA while incurring lower costs. Also, this work finds that biodegradability criteria (natural or industrial) are instrumental in evaluating bioplastics, emphasizing the criticality of end-of-life considerations.Originality/valueThis work is the first to integrate SMAA and CODAS into a unified stochastic multi-criteria decision-making framework and applies it to pioneer a comprehensive evaluation of bioplastics for shopping bags.
Purpose This study proposes a novel hybrid artificial neural network (H-ANN) framework, inspired by reinforcement learning (RL), to proactively detect Internet connection speed problems using enriched datasets from multiple sources of an Internet service provider. Design/methodology/approach The problem is challenging due to the high dimensionality, unbalanced class distribution and continuous influx of new data. To address these issues, the proposed hybrid framework integrates supervised learning methods – radial basis function network (RBFN) and multi-layer perceptron (MLP) – with the unsupervised self-organizing map (SOM). RL is employed to accelerate learning, reduce feature and instance space complexity and improve the detection of underrepresented classes. The framework is first validated on benchmark open-source datasets and subsequently applied to real-world company databases combining network, business and customer information. Findings The results demonstrate that the proposed H-ANN significantly improves both classification accuracy and computational efficiency compared to conventional machine learning approaches. Importantly, the framework enables the early identification of slow Internet connections before customers submit complaints, allowing the service provider to take proactive measures. Originality/value The proposed H-ANN framework not only enables the early identification of slow Internet connections before customers submit complaints – allowing service providers to take proactive measures – but also offers a generalizable solution for large-scale, imbalanced and dynamic data classification problems across diverse domains.
Purpose This study aims to develop an optimized warehouse management systems (WMS) performance evaluation framework by addressing uncertainties such as inventory accuracy, demand variability, risk and resilience, regulatory changes, cost efficiency, environmental control, and technological advancement, ensuring operational efficiency and long-term sustainability. Design/methodology/approach The theory of circular intuitionistic fuzzy numbers (CIFNs) offers a comprehensive framework for identifying contradictory statements and streamlining content evaluation in multi-criteria decision-making (MCDM). The importance of criteria is objectively evaluated using the Method Based on the Removal Effects of Criteria (MEREC), while subjective weights are determined through the Step-Wise Weight Assessment Ratio Analysis (SWARA) technique. These objective and subjective weighting methods are useful for the importance and relevance of criteria in the decision-making process. Findings Alternative rankings are obtained using the AROMAN method with two-step normalization. The proposed MEREC–SWARA–AROMAN algorithm effectively supports WMS performance optimization by simplifying criteria selection and improving decision reliability. Research limitations/implications The computational complexity increases with a large number of criteria and alternatives, limiting scalability for very large datasets. The framework primarily provides static evaluations and may not fully capture dynamic or time-dependent decision environments. Practical implications The framework provides decision-makers with a structured and transparent tool for evaluating WMS performance. It enhances the reliability of alternative ranking, leading to better resource allocation and operational planning. The model is flexible and can be adapted to other real-world decision problems. Social implications The proposed MEREC–SWARA–AROMAN framework offers significant social benefits by enabling more transparent, inclusive, and equitable decision-making processes. By integrating both objective weighting (MEREC) and expert-driven subjective evaluation (SWARA), the model encourages stakeholder participation and reduces bias, thereby enhancing fairness and trust in decisions. Furthermore, its capability to incorporate social criteria—such as social responsibility, public welfare, and stakeholder satisfaction—supports sustainable development goals and improves societal outcomes. Originality/value Unlike prior studies focused on centralized systems, this research presents a real-world MCDM application for big data analysis under fuzziness and vagueness. The proposed framework demonstrates practical effectiveness in decision optimization, with applicability to sustainable energy projects supporting sustainable development.
PurposeThis study explores the external and inherent drivers which motivate mobile banking users to change from menu-based navigation to AI-powered voice and text command chatbots for financial transactions.Design/methodology/approachThe research employs partial least squares structural equation modelling, complemented by Importance-Performance Map Analysis and Necessary Condition Analysis as the main analyses to examine survey data from 230 mobile banking users across the Gulf Cooperation Council region.FindingsThe model employed explains 77.8% of the variance in intentions to change to AI-based Chatbots. It was found that the most powerful motivators were hedonic motivation, habit formation and trust. Hedonic motivation and habit directly affected intentions to change and also influenced ideas about the potential effects of performance and effort expectancies. Expectation of improved performance was an underlying assumption, but could not be used to anticipate changed behaviours, while the prospects of improved workload and security were present, but were not primary determinants, suggesting their role as bottlenecks in adoption. Findings also identified a connection between banking security, hedonic motivation, and habit, and indicated that the perceived banking security reinforced the association between hedonic motivation and habit, thereby increasing the possibility of long-term adoption. Together, the findings clarify that all factors are key and necessary in enabling more user preferences from menu-based mobile banking interfaces to voice and text-enabled AI Chatbots in performing banking chores.Practical implicationsThe results provide useful guidelines for banks and financial institutions who want to motivate customers to adopt AI-based Chatbots. The findings of the study show that banks' actions cannot only focus on performance potential, but also primacy must be given to ensuring trust, providing a pleasurable experience and cultivating habit formation, assuring that the latter two aspects are underpinned by a commitment to security. By fostering digital transformation in banking, these strategies align with the United Nations' Sustainable Development Goal 8.Originality/valueThis study both builds on and addresses gaps in the existing scholarship by considering the important, but still undeveloped topic of why users decide to change their digital banking behaviours to the use of AI Chatbots. Previous studies have tended to focus on intrinsic factors which motivate this change; this research evaluates how these inherent factors mediate the impact of external factors on intentions to make the change to AI Chatbots. Furthermore, this research draws attention to the role of trust and banking security in this decision-making process. This is an area which is not fully addressed in the existing literature. Methodologically, the study extends UTAUT2 by integrating necessity logic and importance-performance perspectives. It is among the first to employ advanced Partial Least Squares Structural Equation Modelling analyses in this setting, thereby contributing novel theoretical and practical insights into mobile banking transformation.
Purpose This study examines the systemic challenges associated with adopting edge–cloud continuum architectures for real-time decision systems and identifies the underlying causal drivers and dependent outcomes shaping sustainable deployment. Design/methodology/approach A Fuzzy Decision-Making Trial and Evaluation Laboratory (Fuzzy DEMATEL) method was employed to model causal relationships among eight key challenges: latency minimization, scalability and data growth, interoperability, cybersecurity and privacy, energy efficiency and cost, reliability and fault tolerance, network bandwidth, and regulatory governance. Pairwise influence assessments were collected from fifteen domain experts and aggregated using triangular fuzzy numbers. Sensitivity analysis was conducted to assess the structural robustness of the causal classifications. Findings The results indicate that latency minimization, scalability, energy efficiency, and reliability function as causal drivers, while governance, interoperability, cybersecurity, and bandwidth constraints emerge as effect-side outcomes. This demonstrates that performance-oriented technical enablers shape downstream institutional and coordination challenges within the edge–cloud continuum. Practical implications The findings offer a decision-support roadmap for managers and policymakers to prioritize upstream technical drivers, enabling more effective mitigation of downstream governance and security risks. Originality/value The study advances soft computing applications by providing a causal, system-level perspective on edge–cloud adoption, moving beyond isolated challenge rankings to reveal indirect influence structures in real-time digital infrastructures.
PurposeThis work aims to investigate and improve adversarial patch attacks for semantic segmentation, a task increasingly deployed in security-critical applications. Existing attacks often overlook pixel-level uncertainty and spatial variation, resulting in inefficient optimization and limited effectiveness. The purpose of this study is to design an uncertainty-aware attack framework that better identifies and exploits structurally vulnerable regions in segmentation models.Design/methodology/approachWe propose a two-stage uncertainty-aware adversarial patch attack framework. The first stage computes pixel-wise entropy to identify locally uncertain regions. The second stage applies a confidence-based inter-pixel weighting strategy that prioritizes vulnerable pixels by comparing their confidence to a global statistical threshold. These components are unified into a dynamic loss reweighting mechanism. Experiments are conducted on Cityscapes and BDD100 K using ICNet, DDRNet, and SegFormer.FindingsExperimental results show that the proposed method outperforms existing patch-based attacks such as SSAP. By effectively targeting uncertain and structurally vulnerable regions, our method achieves stronger degradation of segmentation performance, with mIoU reduced to as low as 8%. The results demonstrate both high attack effectiveness and strong cross-dataset and cross-model generalization.Originality/valueThis work is the first to incorporate pixel-level uncertainty into adversarial patch optimization for semantic segmentation. Unlike prior patch-based attacks that treat all pixels uniformly, our method explicitly models local entropy and confidence-driven spatial variation, enabling more targeted and effective perturbation. The proposed dynamic loss reweighting framework provides a novel perspective on exploiting structural vulnerabilities in dense prediction tasks. This approach offers both theoretical and practical value for understanding segmentation robustness and designing stronger uncertainty-guided attacks.
PurposeFaced with the structural employment contradiction (background) caused by the growth of more than 15% in the gross enrollment rate of higher education and the acceleration of technological iteration in the process of globalization, this paper aims to develop an intelligent system that can accurately predict the employment path of fresh graduates and solve the defects of traditional methods in data imbalance, dynamic adaptability and privacy protection (purpose).Design/methodology/approachBy constructing a hypergraph network based on the relationship between dormitories and courses, this paper innovatively integrates pseudo-label strategy (reinforcement learning to screen real nodes), triple data enhancement (feature masking/hyper-edge modification/graph diffusion) and heterogeneous encoder contrast learning method (L = 2 local layer cooperates with L = 4 global layer).FindingsIn addition, through experimental verification, this paper finds that the model achieves 89.1% accuracy rate and 86.5% minority recall rate in four types of employment status prediction, and the noise robustness is improved to-8.2%. At the same time, the actual deployment has increased the accuracy of entrance prediction by 19% and the enterprise matching degree by 17%, and multi-modal fusion (cross-modal comparative learning) has further pushed the accuracy to 95.1% (result). Therefore, despite the limitations of the model such as domain dependence (the accuracy rate of art migration is 71.3%), a 30% increase in computing power costs, and unstructured data privacy risks.Originality/valueThe model still provides a data-driven solution for solving global "skill misalignment".
