Assessment of normal and abnormal behaviour of people with dementia in living environment through non-invasive sensors and unsupervised AI

The objective of this work is to define a measurement protocol to compute and distinguish abnormal from normal behavior of older people with early to middle stage dementia living alone at home using training artificial intelligence (AI) algorithms (in specific K-means, Agglomerative and Spectral Clustering Algorithms). Unlabeled activities of daily living (ADLs) databases were acquired from an AI-based sensor network composed of three Passive Infrared (PIR) motion sensors and two door sensors which are installed in voluntary participants' houses in Italy, Netherlands, Switzerland and Norway. Our blind approach proposes an unsupervised learning-based solution, representing real-world use case where it is not possible to provide accurate labeling and annotation of the sensor data. Results indicate two out of three algorithms, which applied for 32 participants during 75 days of recording, are valid to create clustering groups for participants with acceptable Silhouette coefficient of 0.77 and 0.65 for Kmeans and Agglomerative clustering, respectively, and 0.13 for Spectral clustering method that makes this algorithm less reliable respect to others.