Analysis of In-Home Movement Patterns for Depression Assessment in Older Adults - A Feasibility Study

Depression significantly impacts the wellbeing of older Australians, posing considerable challenges to their overall quality of life. This study aimed to detect in-home movement patterns of participants that could be indicative of depressive states. Utilising data collected over a 12-month period via smart home ambient sensors, this feasibility study conducted a comparative analysis using machine learning techniques on features derived from motion sensors, sociodemographic variables, and the Geriatric Depression Scale. Three machine learning models, specifically Extreme Gradient Boost (XGBoost), Random Forest (RF), and Logistic Regression (LR), were implemented. Results showed that the performance of XGBoost was relatively higher compared to RF and LR, with an Area Under the Receiver Operating Characteristic Curve (AUROC) value of 0.67. Feature analysis indicated that bathroom and kitchen movements and the level of home care support were among the top influential features influencing depression assessment. This is consistent with clinical evidence on appetite, hygiene, and overall mobility changes during depression. These findings underscore the feasibility of leveraging in-home movement monitoring as an indicator of health risks among older adults.