The United States Air Force seeks to develop “Collaborative Combat Aircraft” to provide an affordable force multiplier. Current Artificial Intelligence (AI) research efforts often focus on developing the AI necessary to support task work within target domains without sufficiently considering integration in existing systems, including the capabilities necessary to enable the teamwork required to enable human-AI teams. The current research employs a recently published team-oriented analysis and design method, referred to as Agent-Oriented Systems Modeling (AOSM), to provide an architecture-level analysis of the human-AI team to establish requirements for the AI systems in both manned and unmanned aircraft. The approach is applied to an F-16 formation performing defensive counter air, to evaluate task and teamwork requirements within the analogous system. These requirements, namely the goals, responsibilities, and capabilities of the agents in the existing system are then applied to envision the “To-Be” system to ensure equivalent or greater capabilities exist in the proposed architecture. This analysis produces numerous traceable findings that can be applied to establish system level requirements for future development to ensure effective human-AI teaming, including 45 capabilities required of the system and five agent roles to be fulfilled in the “To-Be” system. The project discusses and supports the importance of efficient communication and coordination mechanisms for all AI agents when teaming with the human pilot.
Self-regulating emotions, thoughts, and behaviours is central to psychological well-being but often difficult to sustain without structured support, particularly under stress or cognitive overload. This study introduces a human-centred digital system designed to support everyday emotional self-regulation through personalised, explainable self-care guidance grounded in established models of emotion regulation and self-regulated behaviour. The system adapts to users’ reported emotional state, cognitive load, time and energy constraints, and personal preferences to generate actionable self-care plans accompanied by natural-language explanations that clarify the system’s reasoning and contrast alternative strategies, supporting user understanding and autonomy. The system is intended for non-clinical, self-care contexts and does not provide diagnosis or treatment. A formative scenario-based evaluation was conducted using three representative non-clinical user archetypes. The generated outputs were independently assessed using a structured rubric, producing mean scenario scores of 4.2–4.8 out of 5. Findings suggest that explainable digital systems can function as cognitive and emotional scaffolding, enhancing reflective engagement and supporting behaviour change in everyday contexts. This work contributes to research at the intersection of behavioural science and technology by demonstrating how explainability and personalisation can be operationalised in supportive digital systems, offering design insights for transparent, adaptive technologies that promote trust, autonomy, and sustained engagement in psychological well-being support.
The proliferation of artificial intelligence (AI) in complex decision environments has intensified scholarly and practical interest in how humans calibrate trust when collaborating with intelligent systems. This paper presents a cross-domain empirical investigation of human-AI collaborative decision-making across climate intelligence and organisational governance contexts. Drawing on a mixed-methods synthesis of 68 primary studies, original survey data from 200 industry professionals, and 30 semi-structured interviews, we examine trust calibration dynamics, performance outcomes, and implementation barriers. Our findings reveal a systematic trust calibration paradox: while 75
Heart failure is a serious cardiovascular condition which should be assessed for risk at the right time. Using clinical and lifestyle data, machine learning can be utilized to identify patients that may be at risk for heart failure. This study compares supervised machine learning techniques for Heart failure risk prediction from primary data obtained from hospitals and diagnostic centres of Comilla City, Bangladesh. A total of 828 patient records, with 14 clinical, demographic, lifestyle, and comorbidity features, are present in the original dataset. 12 duplicate records were removed, resulting in 816 unique records for model development and evaluation. The statistical feature selection methods used were Pearson correlation and Mutual Information (MI), and the metaheuristic feature selection methods evaluated were Binary Particle Swarm Optimization (BPSO) and Binary Grey Wolf Optimization (BGWO). The classifiers used were compared: Random Forest, Support Vector Machine, K-Nearest Neighbors, Naive Bayes, Decision Tree and XGBoost. Random Forest achieved 96.00
Plantar-pressure analysis is central to contemporary gait research, yet existing reviews typically examine only isolated dimensions, such as sensor engineering, in-shoe system architectures, or disease-specific applications, without providing an integrated perspective. This systematic review addresses this gap by unifying technological advancements, hardware and sensing architectures, clinical applications, and analytical methodologies into a cohesive cross-disciplinary framework. A structured search of IEEE, PubMed, Scopus, and Web of Science (2012–2025), conducted according to PRISMA guidelines, identified 92 studies meeting predefined criteria related to wearable or in-shoe plantar-pressure measurement, sensor-based gait assessment, and clinically oriented mobility evaluation. Through comparative analysis, systems are categorized by hardware configuration, pressure-sensing modality, data-acquisition strategy, and functional scope, clarifying their performance characteristics, constraints, and suitability for real-world deployment. Clinical applications are synthesized across major neurological, musculoskeletal, and metabolic gait disorders, emphasizing shared and condition-specific alterations in plantar-pressure distribution rather than focusing on individual diseases. Analytical methodologies span classical machine-learning pipelines based on engineered features to advanced deep-learning models capable of extracting hierarchical spatial–temporal representations from raw pressure data. Persistent challenges, including sensor drift, low spatial resolution, environmental variability, dataset limitations, and restricted model generalizability, are critically examined alongside practical barriers to clinical translation. The review concludes by outlining future directions involving flexible next-generation sensors, standardized data protocols, interpretable and personalized algorithms, multimodal fusion, and real-time on-device intelligence. Collectively, these insights provide a comprehensive foundation for advancing reliable, scalable, and clinically meaningful plantar-pressure assessment systems.
