Clustering-Based Detection of Unusual Daily Patterns in Elderly Care Using Non-Intrusive Sensor Data

The rise of the global aging population has introduced serious safety challenges. Older adults, while more vulnerable, deserve protection without compromising their independence. Existing Ambient Assisted Living (AAL) solutions often rely on supervised methods, wearable devices, or cloud-based systems that raise privacy concerns, lack scalability, or depend on labeled data. To address these limitations, this paper presents an intelligent, edge-based framework that detects abnormalities in daily behavior using non-intrusive motion and contact sensors deployed in smart home environments. We introduce an unsupervised clustering architecture deployed at the edge, which classifies normal and abnormal days based on unusual activity patterns. The system semantically labels daily routines, enabling interpretable and privacy-preserving anomaly detection. Experiments on a public smart home dataset demonstrate the system’s ability to identify meaningful behavioral clusters and detect deviations, supporting real-time, proactive monitoring in longterm elderly care settings.