Bridging Data and Behavior in Homecare: Personalized Routine Modelling and Anomaly Interpretation

Understanding how older adults organize their daily lives is crucial for developing person-centered homecare systems. This study proposes an interpretable framework for modeling daily life and detecting behavioral anomalies using data from 18 CASAS smart homes. The dataset contains several weeks of continuous sensor recordings from residents living independently. Daily activity patterns were analyzed in 15-minute intervals using principal component analysis (PCA) to identify key temporal patterns shared by the population. For each resident, a personal baseline routine was defined as the median of their daily activity profiles over a 14-day baseline period, and deviations from this baseline were compared with global deviations derived from the PCA model. The results revealed explainable behavioral differences among residents and highlighted three lifestyle archetypes like active bimodal, stable routine, and early resting. By linking the difference scores to contextual activities such as sleep, hygiene, and computer use, the framework provides relevant explanations for daily irregular behaviors.