A Machine Learning-Based Risk Prediction Model for Wandering Dementia Patients

In aging societies, dementia has emerged as a major health challenge accompanied by cognitive decline and loss of daily autonomy, highlighting the growing need for objective monitoring of behavioral changes over time. This paper proposes a behavioral risk analysis framework based on GPS-based mobility data collected at 30-minute intervals over approximately 150 days from a single dementia patient. Three behavioral indicators―hazardous area access (Danger), adverse weather (Weather), and wandering (Wander)―were defined and integrated to compute a behavioral risk score reflecting their spatiotemporal interactions. Experimental results show a gradual increase in risk scores, persistent wandering under adverse weather conditions, increased access to hazardous areas, and repetitive spatial movement patterns, suggesting a progressive decline in environmental awareness and spatial cognitive abilities. These findings demonstrate that GPS-based behavioral data can serve as an objective basis for early detection of behavioral deterioration in dementia patients. Through a single-case study, this work highlights the potential of a structured risk analysis framework integrating long-term behavioral and environmental data, with future applicability to multi-patient datasets and real-time dementia care systems.