Personalized Behavioral Abnormality Detection in Smart Homes

Due to the increasing aging population, the number of people affected by neurodegenerative diseases is expected to grow in the coming years, causing a high cost of elderly care. However, early detection of these diseases can slow down the deterioration of patients’ conditions. This paper focuses on detecting behavioral abnormalities through continuous monitoring of the daily activities of elderly people in smart homes. Human Activity Recognition (HAR) is an area that has been extensively explored in the past few years. However, there is a lack of focused work that leverages AI-driven techniques to identify unusual behaviors due to neurological disorders. In this work, we propose a framework that uses a novel deep-learning sequential model for predicting daily activities using smartphone data and an ontology-based behavioral abnormality detection system. Our knowledge-driven technique caters to multiple abnormal behavioral symptoms related to Alzheimer’s disease and can incorporate any updates in the daily schedule of end users. We use the latest MARBLE dataset (released in 2021) for multi-occupant scenarios and validate our solutions using multiple datasets. Our personalized HAR model is able to achieve accuracy up to 96% even with a new user and is capable of detecting a number of behavioral abnormalities using a rule-based engine. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.