Recent studies on network intrusion detection using deep learning primarily focus on detecting attacks or classifying attack types, but they often overlook the challenge of attributing each attack to its specific source among many potential adversaries (multi-adversary attribution). This is a critical and underexplored issue in cybersecurity. In this study, we address the problem of attacker attribution in complex, multi-step network attack (MSNA) environments, aiming to identify the responsible attacker (e.g., IP address) for each sequence of security alerts, rather than merely detecting the presence or type of attack. We propose a deep learning approach based on Transformer encoders to classify sequences of network alerts and attribute them to specific attackers among many candidates. Our pipeline includes data preprocessing, exploratory analysis, and robust training/validation using stratified splits and 5-fold cross-validation, all applied to real-world multi-step attack datasets from capture-the-flag (CTF) competitions. We compare the Transformer-based approach with a multilayer perceptron (MLP) baseline to quantify the benefits of advanced architectures. Experiments on this challenging dataset demonstrate that our Transformer model achieves near-perfect accuracy (99.98%) and F1-scores (macro and weighted ≈ 99%) in attack attribution, significantly outperforming the MLP baseline (accuracy 80.62%, macro F1 65.05% and weighted F1 80.48%). The Transformer generalizes robustly across all attacker classes, including those with few samples, as evidenced by per-class metrics and confusion matrices. Our results show that Transformer-based models are highly effective for multi-adversary attack attribution in MSNA, a scenario not or under-addressed in the previous intrusion detection systems (IDS) literature. The adoption of advanced architectures and rigorous validation strategies is essential for reliable attribution in complex and imbalanced environments.

Gait is a fundamental and functional activity among older adults; however, aging leads to anatomical and functional gait deterioration. This decline can be mitigated through regular physical activity. Biomechanical analysis—using electromyography, kinematics, and force measurements—offers one of the most objective methods for assessing gait. In this study, we propose a novel framework based on machine learning techniques to identify biomarkers that more precisely distinguish the gait patterns of young adults from those of physically active older adults. Gait analysis included kinematic, kinetic, and surface electromyography (sEMG) data, all recorded under controlled walking speed conditions. The extracted features comprised joint kinematics of the pelvis, hip, knee, and ankle; ground reaction forces (GRF); and muscle activation signals from six muscles in the dominant lower limb. Among the evaluated models, the proposed Multilevel XGBoost approach achieved the highest performance, improving classification accuracy by 13.1%, reaching 80.24%. Key biomarkers identified included pelvic tilt adjustments, reduced ankle range of motion, and altered muscle activation patterns—changes associated with stability-related adaptations in the aging gait. These findings highlight that biomechanical changes are detectable even among older adults who maintain regular physical activity. Future research will aim to integrate deep learning and fuzzy logic techniques to enhance feature extraction and improve the analysis of gait variability.

DEFCON, the world's largest cybersecurity conference, hosts a highly competitive “Capture the Flag” (CTF) competition, renowned for being one of the longest and most challenging in the cybersecurity community. This event typically spans three days and features the top 20 teams globally, each tasked with defending their systems while attacking others. During these events, teams have limited time to patch their services and develop exploits before engaging with other teams to capture their flags. The fast-paced nature of DEFCON CTF events means that success often hinges on the team's experience and agility. Teams face concurrent attacks from multiple opponents and have constrained resources, making it impossible to address all threats simultaneously. In this work, we propose leveraging long short-term memory (LSTM) neural networks to predict the next steps of concurrent multi-stage and multi-step network attacks (MSNAs). This approach aims to enhance team performance by enabling informed decision-making and efficient resource allocation. We extracted attack data from pcap files of CTFs provided by DEFCON, encompassing approximately 300 iterations per CTF. The model was trained on 80% of the initial iterations and validated on the remaining 20%, where more sophisticated behaviors and refined strategies are anticipated. Our methodology achieved a prediction accuracy over 80%, significantly improving response strategies and allowing teams to prioritize threats effectively.

Objectives To report the development and early formative, user-centered evaluation of a human-centric explainable artificial intelligence (AI)-enabled platform for remote and hybrid phase II cardiovascular rehabilitation (CR), and to discuss its policy and technology implications for adoption and governance in health systems facing access constraints. Methods A four-stage methodology was applied: (1) multidisciplinary needs elicitation with cardiovascular rehabilitation professionals; (2) development of machine-learning models for rehabilitation-related risk assessment with integrated explainability; (3) adaptation of expla-nations to clinicians’ and patients’ mental models; and (4) system implementation followed by early multidisciplinary evaluation focused on usability, perceived clinical utility, and safety positioning as a second-opinion decision support tool. Results The platform integrates remote patient monitoring, explainable risk assessment, and coordinated multidisciplinary workflows. In early formative evaluation, healthcare professionals reported high acceptance of the explainable second-opinion functionality, highlighting improved interpretability and support for rehabilitation assessment and discharge-related discussions, without replacing clinical judgment. Conclusions This study provides an early-stage, policy-relevant account of how explainable AI can be operationalized in cardiovascular rehabilitation while remaining aligned with clinical practice and governance expectations. Rather than demonstrating system-level impact, the con-tribution lies in outlining a practical framework for evaluating adoption conditions, governance needs, and future scale-up of AI-enabled rehabilitation technologies. Public interest summary Cardiovascular rehabilitation helps people recover after a heart event, but many patients face barriers to attending in-person programs, particularly due to distance, mobility, or limited service availability. SITeCard is a digital platform developed to support remote and hybrid cardiovascular rehabilitation by organizing patient data and providing clinicians with explainable AI-based risk assessments to inform multidisciplinary discussions. The system was co-designed with healthcare teams to ensure usability and clinically meaningful explanations, and it can be accessed through standard smartphones, including in low-connectivity settings. This study reports early, user-centered evaluation results and highlights policy and governance considerations relevant to the adoption of explainable AI tools in rehabilitation services. Graphical abstract Conceptual overview of SITeCard, an early-stage, human-centric explainable AI-enabled platform designed to support remote and hybrid phase II cardiovascular rehabilitation as a second-opinion clinical decision support system. The platform integrates patient data, explainable risk assessment, and multidisciplinary workflows. Patient and system outcomes are shown as policy-relevant targets to be validated in future large-scale evaluations.

The rapid adoption of Machine Learning (ML) in high-impact domains has intensified the need for systematic tools to assess and improve the trustworthiness of predictive models beyond conventional performance metrics. This paper presents TWEEF (Trustworthiness Estimation and Enhancement Framework), a modular and extensible framework that operationalizes trustworthiness through the joint evaluation of performance, fairness, and interpretability. TWEEF integrates intuitionistic fuzzy logic and subjective logic to transform quantitative trust-related metrics into linguistic assessments, which are subsequently aggregated using operators such as the Linguistic Weighted Average (LWA), Gaussian Weighted Aggregation (GWA), and Subjective Logic (SL). The framework extends the scikit-learn ecosystem through a meta-estimator, the TrustworthyClassifier, which orchestrates metric computation, bias-mitigation procedures, surrogate-model generation, and trust aggregation within a unified, pipeline-compatible workflow. The framework is empirically evaluated through four experiments on widely used benchmark datasets (German Credit, COMPAS, and Adult) in binary classification settings. Results show that TWEEF consistently reveals fairness and interpretability limitations that may remain hidden when relying solely on predictive performance, and that the resulting trust scores respond coherently to different metric configurations and weighting schemes. These findings indicate that TWEEF provides a structured mechanism for trust assessment and enhancement, while also offering a flexible foundation for future extensions to additional learning tasks and evaluation dimensions.