An integrated AI-IoT framework for elderly mental health monitoring: architecture and offline feasibility evaluation
Global population ageing is intensifying the demand for scalable approaches to elderly mental health support. Existing care models remain largely episodic and insufficient for continuous monitoring, concurrent at-risk status assessment, and coordinated response. This paper presents the AI-Integrated Mental Health Support System (AIMHSS), a layered framework that combines Artificial Intelligence, Internet of Things, Virtual Reality, and blockchain technologies to support elderly mental health monitoring and intervention planning. The framework comprises four functional layers, namely Continuous Sensing, Predictive Intelligence, Adaptive Intervention, and Stakeholder Engagement, supported by cross-cutting trust, data integrity, and ethical governance mechanisms. Technical feasibility was examined through a two-part offline evaluation using two publicly available datasets containing authentic observational data. First, a Random Forest classifier of concurrent at-risk status was trained and evaluated on the OASIS-2 longitudinal clinical dataset under leakage-safe subject-isolated validation, achieving an AUROC of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.823 \pm 0.038$$\end{document}, a sensitivity of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.664 \pm 0.079$$\end{document}, and a specificity of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.845 \pm 0.090$$\end{document}. Second, Fitbit wearable activity and sleep traces were used for trace-driven orchestration testing with a fixed synthetic clinical baseline, enabling behavioural monitoring and tiered intervention triggering across a 12-user cohort comprising 331 user-days. These evaluations establish architecture-level and pipeline-level feasibility for integrated elderly mental health support and define a clear next step of prospective multi-modal validation within unified clinical cohorts.