A Hierarchical Rule-Based Activity Recognition System with Frequency Pattern Mining

In this paper, a hierarchical system is proposed for generating personalization action and activity recognition rules. Multi-level decision rule mining approach in our system not only discovers personal habit of device using, but also finds personal pattern of devices operation manner. First, it processes non-sequential procedure, which mines user´s historical action database for constructing rules to recognize actions. Second, it creates action sequence recognized by using action rules generated before. Sequential mining method is adopted to find out the most frequent sequential pattern which is seemed as personal habit of device operating manner. Finally, non-sequential rules and sequential patterns are checked to recognize user´s activity. The system generates rules depending on user´s habit rather than being specified by designer who only creates general rules. The simulation results reveal that the activity recognition accuracies with 10% and 20% noise interference are 87.33% and 83.33% respectively.