Predicting activities of daily living via temporal point processes: Approaches and experimental results

Activity Prediction is foreseeing the following activity people are going to execute. This is a crucial task in smart home environments, i.e., in order to facilitate the daily routines of elderly people with or without special needs. In this paper, we focused on Activity Daily Living prediction and we proposed a novel activity prediction technique based on the combination of Marked Temporal Point Processes and Neural Networks. Experiments on real and synthetic smart space datasets have shown that our approach is able to conveniently represent and predict daily living activities in an unsupervised way. We evaluated its performance and compared its results with state-of-the-art methods providing freely available implementations. Noticeably, the proposed approach outperforms the best concurrent algorithm by obtaining an improvement of F1-score of 60% (on average of the considered datasets).