Unsupervised statistical concept drift detection for behaviour abnormality detection

Abstract Abnormal behaviour can be an indicator for a medical condition in older adults. Our novel unsupervised statistical concept drift detection approach uses variational autoencoders estimating the parameters hypothesis test abnormal days. As feature, Kullback–Leibler divergence of activity probability maps derived from power and motion sensors were used. We showed general feasibility (min. F 1 -Score 91 %) on artificial dataset four types. Then we applied our new method to real–world collected homes 20 (pre–)frail adults (avg. age 84.75 y). was able find days when participant suffered severe condition.