A novel approach for discovering human behavior patterns using unsupervised methods

Abstract

As is well known, elderly people gradually lose the ability of self-care. The decline can be reflected in changes in their daily life behavior. A solution to assess their health status is to design sensor-enhanced living environments to observe their behavior, in which unobtrusive sensors are usually used. With respect to information extraction from the dataset collected by means of these kinds of sensors, unsupervised methods have to be relied on for practical application. Under the assumption that human lifestyle is associated with health status, this study intends to propose a novel approach to discover behavior patterns using unsupervised methods. To evaluate the feasibility of this approach it was applied to datasets collected in the GAL-NATARS study. The study is part of the Lower Saxony research network Design of Environments for Aging (GAL) and conducted in subjects' home environments. The subjects recruited in GAL-NATARS study are older people (age a parts per thousand yenaEuro parts per thousand 70 years), who are discharged from hospital to live alone again at their homes after treatment of a femoral fracture. The change of lifestyle regularity is measured. By analyzing the correlation between the extracted information and medical assessment results of four subjects, two of them exhibited impressive association and the other two showed less association. The approach may provide complementary information for health assessment; however, the dominant relationship between the change of behavior patterns and the health status has to be shown and datasets from more subjects must be collected in future studies. Merely environmental data were used and no wearable sensor for activity detection or vital parameter measurement is taken into account. Therefore, this cannot comprehensively reflect reality.