Applying Hierarchical Information with Learning Approach for Activity Recognition

Abstract

Listen

Translate article

This paper discusses the problem of applying ontology for activity recognition and proposes a hierarchical classification approach by using categorize information among activities with machine learning method. In activity recognition problem, machine learning approaches have the ability to adapt to various real environments but actual setting do not often obtain enough quality data to construct a good model for recognizing multiple activities. Our approach exploits the hierarchical structure of activities to overcome the problem uncertainty and incomplete data for multi-class classification in real home setting datasets. While slightly improves the overall recognition accuracy from 59% to 63%, hierarchical approach can recognize infrequent activities such as “Going out to work” and “Taking medication” with accuracies of 80% and 56% respectively. Those activities had recognition accuracies lower than random guess in previous learning method. The preliminary results support the idea to develop a methodology to utilize semantic information represented in ontologies for activity recognition problem.KeywordsRecognition AccuracyActivity RecognitionHierarchical ApproachPersonal HygieneCategory LabelThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.