Abnormal Behaviour Detection in Smart Home Environments

The study of user behaviour patterns in activities of daily living is significantly challenging due to the ambiguity in identifying a usual and unusual activities. Capturing the user activities using ambient sensors in a smart home environment serves as the safest and most effective way. Detecting the unusual activities of the residents helps in regularizing their daily tasks. The paper focuses on unsupervised methods for anomaly detection based on clustering. The comparison study is done between different unsupervised clustering models: self-organizing map, density-based clustering, and Gaussian mixture models. The unsupervised clustering recognizes the normal user behaviour of each activity and looks for any deviations or abnormalities. The experimental study is done on real-time public datasets, and the silhouette score were used to identify the best abnormality rate for each activity.