Search Results (101)

Phishing email remains a persistent and operationally critical cybersecurity threat, yet existing detection approaches, including traditional machine learning and single-pass large language model systems, either lack native interpretability or provide explanations that are difficult to standardize and audit. This paper introduces an explainable multi-agent LLM framework that decomposes phishing evidence across three role-specialized agents focused on linguistic patterns, psychological manipulation, and sender identity consistency. The framework then aggregates specialist outputs through schema-governed synthesis, enabling each intermediate and final decision to be structured, comparable, and auditable. The central contribution is the treatment of role-specialized evidence decomposition and explanation structure as first-class design constraints rather than post hoc additions. The framework is evaluated on a fixed 1000-email subset drawn from a unified TREC/Nazario corpus of 56,212 emails under controlled zero-shot conditions. The full multi-agent Meta-Judge system achieves Macro-F1 = 98.28% and phishing recall = 99.45%, improving Macro-F1 by 6.3 percentage points over a zero-shot single-model GPT-4o-mini baseline. Paired statistical testing confirms that this improvement is significant and is driven primarily by reduced false positives on legitimate emails while preserving high phishing recall. Additional evaluation on an independent LLM-attributed email benchmark shows a consistent Macro-F1 improvement of 0.0773 over the zero-shot baseline under distribution shift. Ablation results show that role-specialized decomposition is the primary performance driver, while deterministic voting provides a competitive raw-classification aggregator and Meta-Judge synthesis provides structured, analyst-facing explanations. These results indicate that role-specialized evidence decomposition combined with schema-governed explanation can improve both detection reliability and auditability in phishing classification workflows. Full article

Existing post hoc XAI (Explainable Artificial Intelligence) methods produce numerical attributions without symbolic structure (SHAP, LIME), low-coverage local rules (Anchors), or crisp tree surrogates without an interpretable rule-level uncertainty proxy. We present a fuzzy rule-based explanation framework for tabular black-box classifiers, extracting global IF–THEN rules with linguistic labels. This was validated on a 13-dataset benchmark with four model families (Wilcoxon, Friedman, TOST equivalence): (i) prediction-boundary-aware fuzzy partitioning raises mean fidelity from a vanilla Wang–Mendel baseline of 0.736 to 0.893 (+10.4 pp excluding the Breast Cancer outlier; +15.7 pp aggregate, both transparently reported); (ii) fired-rule consequent entropy provides a zero-cost rule-level uncertainty proxy (Spearman ρ = 0.420 with model prediction entropy, significant on 11/12 datasets—moderate by Cohen’s convention, with a 4/12 weak-correlation tail; complementary to probability-entropy and margin baselines). Fidelity is statistically equivalent to tree surrogates on classification (TOST p = 0.002, δ = 0.05) at ≈100% coverage. SHAP/LIME are excluded from the formal stability ranking because the perturbation metric measures the wrapped black-box rather than the attribution vector; cross-explainer comparison is reported in grouped form (full-coverage surrogates vs. local-coverage methods). On continuous regression (California Housing fidelity 0.422 vs. TreeSurrogate 0.840) and XOR-type multi-feature interactions, the framework is structurally weaker, addressed by a planned TSK extension. Full article

Crash severity prediction is critical for emergency response, infrastructure spending, and risk communication. Although machine learning has been widely applied to this problem, three gaps prevent practical deployment: uncalibrated probability scores, SHAP-based explanations whose faithfulness has not been verified, and models never tested outside their training jurisdiction. The proposed framework, SAE-XCrash (Safety-Aware and Explainable Crash Severity Prediction), addresses all three using two public datasets—US-Accidents (7.0 million records, 2016–2023) and UK STATS19 (approximately 1,010,000 records, 2016–2022)—with strict temporal splits throughout. Notably, the US-Accidents severity label measures traffic disruption duration, not injury outcome; results should be interpreted accordingly. Previously unknown label-schema drift led to a revised binary target with Severity 4 as the only positive class. Five classifiers are compared. Post hoc isotonic calibration reduces Expected Calibration Error by 97.3% at negligible discrimination cost. A four-step quantitative SHAP audit confirms statistically significant deletion faithfulness; however, explanation stability fails at realistic perturbation levels (54.3% low-stability fraction at sigma = 0.05), driven by spatial data sparsity in sparse geohash cells—a negative result that carries direct operational implications for deployment. A three-tier cross-dataset transfer experiment (zero-shot, recalibration, full retrain) shows that temporal features transfer robustly across jurisdictions, while spatial memorization is the primary generalization barrier. All code, split indices, and model artifacts are publicly available. Full article

Intelligent autonomous driving systems must not only predict the appropriate driving manoeuvre but also provide human-interpretable evidence that justifies the decision. However, existing methods typically address these objectives separately, leading to three practical limitations: multi-stage perception-to-language pipelines can propagate upstream perception errors into downstream explanations; post hoc saliency methods often produce pixel-level highlights that are difficult to interpret semantically; and decoupled decision and explanation modules cannot guarantee that the explanation reflects the same scene evidence used for behaviour prediction. In this paper, we propose a unified framework that jointly performs vehicle behaviour prediction and human-centric interpretation from a shared visual backbone. Specifically, a hierarchical Swin Transformer encodes the driving scene into a sequence of spatial tokens, which are processed by two complementary branches. The first branch, termed the Object Selection Module (OSM), learns a compact scene-level semantic representation through query-guided cross-attention, while the second branch extracts a small set of class-agnostic object-centric tokens without requiring bounding-box or segmentation supervision. These two representations are subsequently integrated by a Semantic–Object Fusion (SOF) module based on scaled dot-product attention, residual connections, and a feed-forward network. The behaviour prediction head operates on the fused representation, whereas the interpretation head leverages the semantic representation through a skip connection to preserve decision-relevant context. For surround-view perception, learnable per-camera embeddings are introduced to maintain viewpoint identity with negligible additional parameter cost. Furthermore, a compact language model fine-tuned via Low-Rank Adaptation (LoRA) generates fluent, label-conditioned natural-language justifications. Extensive experiments on two public benchmarks, BDD-OIA and nu-AD, demonstrate that the proposed framework consistently delivers superior performance and provides effective, human-readable interpretations of driving decisions. Full article

Rare-disease diagnosis is difficult because clinicians must identify plausible conditions from a large, severely imbalanced disease space using evidence distributed across clinical narratives, structured findings, and image-linked descriptions. This paper presents a hybrid pipeline with caption-mediated multimodal fusion for ranked rare-disease diagnosis and grounded explanation, developed and evaluated on the ZebraMap multimodal case-report dataset (69,146 structured cases; 1727 diseases). Grouped train–validation–test splitting by source article was applied to prevent leakage, and a sequential pipeline was constructed combining BM25 lexical retrieval, a class-balanced TF–IDF classifier, MedCPT dense retrieval and cross-encoder reranking, caption-based image-aware late fusion, and post hoc grounded explanation generation. The final pipeline achieved test MRR 0.3905 and Recall@10 0.5507 (nDCG@10 0.4273), while the strongest individual component, the class-balanced TF–IDF classifier, reached MRR 0.4200 and Recall@10 0.6279; the hybrid pipeline therefore integrates ranking with grounded explanation rather than maximizing single-metric diagnostic accuracy. On 256 explained cases, the explanation module achieved citation coverage 0.7334 and usefulness 3.8734, exposing a tradeoff between diagnostic accuracy and explanation richness. These results indicate that a hybrid retrieval-and-classification approach can support ranked rare-disease differential diagnosis and that grounded explanation quality can be evaluated quantitatively, extending computational support for the prolonged rare-disease diagnostic process. Full article

The concept of Elastic Patterns was originally proposed as a prototype-based classification approach that unifies perspectives from cognitive psychology, fuzzy logic, and physics. At their core, Elastic Patterns operate across two levels of deformation: a parameter-level deformation, quantified in terms of axial strain, and a pattern-level deformation, understood as the accumulation of deformation energy to perfectly fit the sample to be classified. This dual representation supports an interpretable and adaptive recognition mechanism, where classification emerges from selecting the Elastic Pattern that requires the minimal deformation energy to align with a real case to classify. This paper extends the theoretical and practical foundations of the proposed Elastic Patterns approach for adaptive pattern classification by introducing several deformation models, Spring Hardening, Weighted Spring Deformation, or Group Parameter Deformation to improve the capacity of Elastic Patterns to adapt to different contexts. These deformation models allow the proposal to adapt to different semantic contexts by controlling how parameter contraction and elongation are penalised. Additionally, novel adaptive mechanisms are introduced, which enable Elastic Patterns to dynamically adjust parameter relevance, capture inter-parameter dependencies, and better reflect contextual knowledge. Furthermore, the framework offers inherently interpretable classification via explicit parameter deformations and energies, avoiding post hoc explanations. Full article

The increasing complexity and distributed nature of Robot Operating System (ROS)-based robotic systems require advanced Fault Detection and Diagnosis (FDD) approaches that operate autonomously with minimal human intervention. The goal of this systematic literature review is to investigate how observability-driven FDD can be automated in ROS-based robotic systems to minimise human effort. Through this lens, the review surfaces four recurring gaps that collectively limit observability-driven automation: rich telemetry sources—logs, traces, and metrics—exist in isolation and are rarely integrated into real-time detection pipelines or leveraged collectively to improve failure diagnostics; online monitoring enables automatic fault detection but depends heavily on predefined rules and expert configuration and interpretation; failure explanations are generated post hoc and rely heavily on logs; and systems remain largely reactive, lacking the continuous monitoring infrastructure needed to anticipate faults before they propagate. Although Large Language Models (LLMs) show considerable promise for automated fault explanation and natural language interaction with robotic systems, current implementations fall short of comprehensive, real-time monitoring that unifies logs, traces, metrics, and sensor streams with Artificial Intelligence (AI) reasoning. To address these gaps, this paper motivates hybrid architectures that combine observability-first design, runtime monitoring, static analysis, and agentic LLM-based reasoning, laying the groundwork for more proactive and autonomous fault management in ROS-based systems. Full article

Multivariate time-series anomaly detection is often evaluated with point-adjusted metrics, which can overstate practical performance when alarms are judged at the event level. Explanation results are also frequently reported as descriptive attributions without directly testing whether selected variables are useful for diagnosis. This study revisits these issues through unified event-level evaluation and repair-based explanation, using DCdetector as the main case study rather than proposing a new detector architecture. Experiments on SMAP, MSL, and HAI 21.03 use full-coverage score export and standard event-level control metrics. The results show that point-adjusted scores can be much higher than stricter event-level measurements. Event-aware refinement changes the detection trade-off by improving event recovery and reducing delay in several settings, but its effect is dataset- and calibration-dependent. For explanation, variables are ranked by exact marginal counterfactual repair effect and evaluated by whether repair reduces anomaly scores more than random or heuristic alternatives. The results provide quantitative evidence that the ranked variables are diagnostically informative, while exact marginal verification is computationally expensive and better suited to offline alarm review and post hoc diagnosis than latency-critical deployment. Auxiliary checks with TranAD, Anomaly-Transformer, and DADA support the plausibility of the main observations, but the evidence remains detector-conditioned rather than a fully backbone-agnostic benchmark. Overall, this work provides a stricter and more verifiable protocol for evaluating anomaly detection, event-aware refinement, and explanation quality in multivariate time-series monitoring. Full article

Student dropout remains a critical challenge in higher education, requiring early detection and targeted intervention. This study aims to develop an interpretable two-level monitoring framework for identifying at-risk students—those with academic debts but not yet dismissed—across successive stages of the academic debt lifecycle. Using digital profile data and LMS digital footprints from a large public university (18,192 records covering the years 2022–2024), we trained CatBoost, XGBoost, LightGBM, and Random Forest for each of two stages: initial retakes and final commission retakes. SHapley Additive exPlanations (SHAP) were applied for post hoc interpretation. SHAP analysis identified key indicators of initial retake failure: semester, year of study, number of academic debts, GPA in the previous semester, and LMS activity in the previous and current semesters. The strongest indicator of success on commissions was the presence of a digital footprint at the beginning of the current semester, which eliminated dropout risk regardless of prior academic history. Dismissal risk increases for junior-year students and those with higher debt counts. These findings enabled student profiling into Red, Yellow, and Green risk categories for optimized allocation of administrative and tutoring resources. Utilizing the proposed framework, educators can streamline pedagogical support and enhance student retention. Full article

The deployment of Deep Learning (DL) for anomaly detection in Industrial IoT (IIoT) is critically hampered by the non-stationary nature of industrial data streams and the lack of forensic-grade explainability. This systematic review synthesizes 48 peer-reviewed studies (2021–2025) to quantify the performance collapse of static models under concept drift and to establish operational criteria distinguishing post hoc feature attribution (Type A XAI) from forensic root-cause diagnosis (Type B XAI). Our analysis reveals three critical findings: (1) static DL models suffer a 15–22% F1-score degradation across wastewater, manufacturing, and energy sectors when deployed in non-stationary environments, rendering them operationally non-viable without continuous adaptation; (2) the current literature remains saturated with Type A explainability (80% of corpus through 2023), creating a Forensic Gap where operators receive statistical correlations but lack actionable maintenance directives; and (3) emerging 2024–2025 research marks a paradigm shift toward Type B methodologies, yet no unified framework bridges real-time detection with deep causal reasoning. To address these gaps, we contribute the following: (1) a validated operational taxonomy (Cohen’s $\kappa =0.84$) with reproducible five-criterion rubric enabling forensic XAI classification; (2) the first quantitative synthesis of drift penalties in industrial deployments; and (3) a three-tier Edge-Cloud Forensic XAI architecture that achieves 70% communication payload reduction via compressed latent vectors while integrating tnGAN-based data imputation (handling 20–30% missing data) and physics-guided causal reasoning engines. Our framework decouples millisecond-level edge detection from 1–3 s cloud-based forensic diagnosis, ensuring both operational responsiveness and actionable industrial insight. We conclude that the future of safety-critical IIoT demands “Forensic-by-Design” architectures leveraging machine unlearning for drift adaptation and LLM-based natural language interfaces for operator-facing explanations, positioning Industry 5.0 to bridge the gap between algorithmic detection and human-centered decision support. Full article

Dropout prediction in Massive Open Online Courses (MOOCs) has been extensively studied, yet existing systems share three fundamental limitations: Accurate models are black boxes, post-hoc explanations approximate rather than faithfully represent model decisions, and predictions are rarely translated into concrete instructor actions. This paper presents NeSy-Drop, a neuro-symbolic framework that simultaneously addresses prediction, explanation, and personalized intervention routing for MOOC dropout. NeSy-Drop constructs a heterogeneous graph per course cohort encoding student–resource–assessment interactions, processed through a heterogeneous graph transformer encoder, five behavioral atom predictor MLPs, and a differentiable symbolic rule layer producing guaranteed faithful ante-hoc explanations. A three-level explainability stack provides symbolic rule chains, SHAP embedding attribution, LIME raw-feature importance, and gradient-based counterfactual prescriptions. Each at-risk student is routed to one of five concrete interventions at one of three severity levels. Evaluated on OULAD covering 32,593 students across 22 cohorts, NeSy-Drop achieves AUC of 0.961 and macro F1 of 0.8983, within 2.2% AUC of the best non-interpretable baseline under a fair evaluation protocol, while being the only system that simultaneously predicts, explains, and prescribes actions at the individual student level. Full article

In industrial machine learning, predictive performance alone is insufficient to ensure reliable deployment, as model behaviour may vary across different regions of the input space under limited data and evolving process conditions. This work investigates whether such variation can be systematically analysed through post-hoc methods. A model-agnostic framework is proposed in which interpretability metrics, including residuals and feature attributions, are embedded into a low-dimensional space and approximated using a continuous surrogate model. This representation enables the analysis of model behaviour as a structured landscape, rather than as isolated pointwise explanations. The approach is applied to ceramic heating element production, where two distinct regimes are identified. One corresponds to a stable region with consistent and accurate predictions, while the other reflects a transitional regime associated with increased ambiguity and sensitivity to feature interactions. These regimes are shown to align with known process conditions and temporal variation. The results demonstrate that model behaviour can be organised into coherent regions that are not observable through aggregate performance metrics alone. This provides a structured basis for post-hoc analysis, supporting targeted interpretation and further investigation of model reliability in industrial settings. Full article

As IoT systems complexity grows, transparent and trustworthy machine-learning intrusion detection systems are crucial. Post hoc explainable AI methods, such as SHAP and LIME, are the most widely used ways to explain how models work, but the degree to which these methods are robust to adversarial conditioning is understudied. In this paper, we propose to create a unified system of evaluating explanation fidelity by using three metrics: sparsity, completeness, and robustness based on minimally distorting DeepFool input perturbations. Our study benchmarks SHAP and LIME across three datasets (BoT-IoT, Edge-IIoT, and N-BaIoT) using four classifiers: CNN, DNN, LSTM, and RF. Our results demonstrate a consistent trade-off: SHAP achieves stronger feature alignment and higher completeness under attack, whereas LIME exhibits greater rank stability in terms of top-k feature overlap. However, LIME also produces more spurious attributions and offers less explanatory power than SHAP, especially in the presence of synthetic features. Our findings reveal that high model accuracy does not guarantee that the provided explanation is also high-fidelity. This investigation highlights the necessity for robustness-aware XAI in cybersecurity and provides reproducible parameters to guide the adoption of XAI in adversarial environments. Full article

Background: Explainable artificial intelligence (XAI) is increasingly proposed to improve trust in mammography-based artificial-intelligence systems, but the validity and clinical readiness of published explanations remain unclear. We aim to systematically review XAI methods applied to mammography and synthesize how explanations are evaluated for validity, robustness, and clinical usefulness. Methods: We conducted a systematic review according to PRISMA 2020. MEDLINE/PubMed, Embase, Scopus, Web of Science Core Collection, and the Cochrane Library were searched from 1 January 2015 to 15 January 2026. Two reviewers independently screened records and extracted data; disagreements were resolved by consensus with a third reviewer. Included studies used mammography as the primary input and reported an explicit explanation or interpretability mechanism. Because the literature was methodologically heterogeneous, we performed a structured narrative synthesis and an adapted XAI-specific appraisal of explanation claims, quantitative evaluation, external validation, human-factor assessment, and reporting transparency. Results: Fourteen studies were included. Ten studies addressed detection or lesion classification and four addressed risk or outcome prediction. Primary XAI families were interpretable-by-design architectures (6/14), post hoc saliency or attribution methods (5/14), and feature-level explanation methods (3/14). Five studies remained at tier-1 qualitative plausibility only, seven reached tier-2 internal quantitative explanation evaluation, two reached tier-3 external or cross-dataset interpretability assessment, and none reported reader or workflow studies. In the dedicated mammography saliency benchmark, Pointing Game scores for Grad-CAM, Grad-CAM++, and Eigen-CAM ranged from 0.30 to 0.41, indicating only modest lesion-pointing reliability despite acceptable classifier performance. Conclusions: Mammography XAI remains dominated by visually plausible explanations that are inconsistently validated. The literature is moving toward task-aligned and intrinsically interpretable designs, yet external validation and clinician-centered evaluation remain rare. Future studies should pre-specify explanation claims, use task-appropriate quantitative metrics, report explanation robustness under distribution shift, and test whether explanations improve human decision-making. Full article

This study proposes a scalable Explainable Artificial Intelligence (XAI)–driven Bayesian–AI decision–control framework for multi-objective optimisation in uncertain and dynamic systems. The framework integrates Bayesian networks, stochastic control, and expected utility theory within a unified probabilistic architecture. Unlike traditional black-box models, the proposed framework provides intrinsic interpretability through probabilistic reasoning and dependency-aware modelling. This allows users to understand how decisions are formed and how variables influence outcomes. To further strengthen explainability, the framework incorporates post hoc XAI techniques, including SHAP-based feature attribution and sensitivity-based local explanations. These methods quantify the contribution of each variable and provide clear explanations at both global and local levels. The system is formulated as a stochastic state-space model and implemented as a closed-loop adaptive architecture. It updates decisions continuously as new data becomes available. Scalable inference is achieved using variational inference, Markov Chain Monte Carlo, and Sequential Monte Carlo methods. This ensures efficient performance in complex and high-dimensional environments. A simulation study based on 370 observations shows that the proposed framework improves decision quality, robustness under uncertainty, and transparency compared to conventional methods. Explainability is evaluated using Fidelity, Stability, and Transparency metrics. The results confirm that the model produces consistent and reliable explanations. The framework supports human-centred decision-making by providing visual analytics and clear probabilistic explanations. This makes it suitable for high-stakes applications such as cyber–physical systems, intelligent platforms, and real-time AI systems. The main contribution of this study is the integration of intrinsic probabilistic interpretability with post hoc XAI techniques into a single, scalable framework. This approach bridges a key gap in XAI research and offers a practical and transparent solution for decision-making under uncertainty. Full article