Machine Learning for Wearable Devices Addressing the Sustainability of an Aging Population

Abstract

Population aging has increased the need for continuous health monitoring, early risk identification, and accessible long-term care. Wearable devices can collect physiological, behavioral, and location data during daily activities, but these measurements require reliable analysis before they can support health-related decisions. This article presents a narrative review of machine-learning applications in wearable health monitoring for older adults. It examines wearable sensing and data-transmission systems, describes an end-to-end framework from signal preprocessing and feature representation to health-state recognition, risk prediction, personalized recommendations, and user or clinical feedback, and compares reported performance across representative applications. Existing studies show that machine learning has been applied to activity recognition, fall detection, arrhythmia classification, stress assessment, rehabilitation monitoring, and personalized health support. However, results vary across datasets, sensor modalities, target populations, and evaluation protocols. The long-term contribution of wearable systems to an aging population therefore depends not only on predictive accuracy but also on signal quality, model interpretability, privacy protection, device reliability, aging-friendly interaction, and appropriate response pathways. Current evidence remains limited by small or heterogeneous samples, cross-device differences, and insufficient longitudinal validation in real-life settings. Further work is needed on individual-baseline modeling, multimodal integration, privacy-preserving learning, cross-device evaluation, and long-term assessment in everyday settings. These developments may support continuous health management, independent living, and more targeted use of healthcare resources among older adults.