Development of an AI-Enabled Remote Loneliness Assessment System for Older Adults Using Behavioral and Speech Data From IoT Sensors and Smart Speakers

Loneliness is a critical risk factor for adverse health outcomes in older adults, yet remains under-detected due to social stigma, time constraints, and reliance on episodic self-report. We describe HomePal, a passive system designed to support future assessment of loneliness-relevant behavioral patterns in older adults living alone. HomePal integrates in-home sensors, smart speakers, and phone communication logs to capture mobility, environmental engagement, sedentary behavior, sleep, TV use, voice characteristics, and communication patterns over 3 months. Because loneliness overlaps conceptually and empirically with related constructs such as depression and social isolation, we emphasize a construct-driven modality selection framework in which each behavioral indicator is treated as inherently ambiguous and interpretation depends on converging evidence across modalities rather than any single channel. This article details the system's development, including participant-centered deployment procedures, methods for translating sensor data into clinically interpretable behavioral patterns, and a planned semi-supervised learning framework linking behavioral indicators to biweekly UCLA Loneliness Scale assessments alongside planned analyses of convergent and discriminant validity. Illustrative case examples demonstrate the feasibility of deriving multimodal behavioral indicators in real-world home environments. We provide a methodological foundation for passive, ecologically valid AI-based loneliness assessment in older adults.