Abnormal Behavior Detection for Elderly People Living Alone Leveraging IoT Sensors

E-health home based solutions reduce healthcare costs and allow aging population to continue their daily life independently. Our objective, is to combine simple IoT sensors and machine learning techniques, in order to provide a home based solution that is able to detect behavioral changes of elderly people who live alone. For this purpose, we introduce a non-intrusive, spatio-temporal abnormal behavior detection approach. In this approach, motion and door sensor signals are elaborated to produce contextual metrics, which are filtered from any deviant observations, after performing a silhouette analysis on five outlier detection algorithms. Next, the combination of a classification and a regression based approach is proposed for detecting abnormalities in the metrics, both in the contexts of space and time. IoT sensor data from ten elderly people houses have been collected and seven different machine learning algorithms have been analyzed in order to evaluate the performance of the individual as well as the combined approach.