Unsupervised Activity Recognition using Temporal Data Mining
Excellent results have been obtained from data mining techniques in many areas. This paper presents one such technique, in the context of activity recognition in a smart home. The authors use sequential pattern mining to analyze the history of information transmitted by the sensors, discovering thereby the frequent activities of the home occupant. Then each of the activities is temporally segmented, in order to facilitate the recognition of activities already started or even ones that are about to start. Their tests revealed that this segmentation diminished the activity search time by more than 70%, and helped predict some activities before detecting any action.