Abstract
Human activity recognition (HAR), i.e., the automated detection and classification of specific activities that a person pursues, is one of the core concerns of mobile and ubiquitous computing. Multimodal sensing facilities of modern mobile devices allow for detailed capture of contextual information, most importantly movement data recorded with inertial measurement units that are now standard in most mobile devices. The majority of HAR applications aim at automatically documenting when something of interest has happened and what that was. For example, the popular moves app on iOS and Android devices "automatically records any walking, cycling, and running [a user does]" [7] and as such automatically generates a life log for those interested in their daily movement patterns. Beyond the mere recognition of certain activities of interest, few applications currently go a step further and analyze the quality of a person's activities, i.e., how (well) their activities were performed, which directly corresponds to a person's abilities or skills .