Performance evaluation of classification methods for online activity recognition on smart phones

Human activity recognition using sensory data has become an active field of research in the domain of pervasive and mobile computing. It involves the use of different sensing technologies to automatically collect and classify user activities for different application domains. In fact, smart phones with their sensing capabilities can also be used as a platform for human activity recognition. Although many studies have been introduced so far, there are few which consider online training and classification of activities as well as evaluating the online performance of existent classifiers. In this thesis, our aim is to analyse the performance of different classification methods for online activity recognition on smart phones using the built-in accelerometers considering important limitations of the phones, such as battery usage and limited computational power. For this purpose, we developed a mobile application on Android platform which performs online classification. We conducted experiments to investigate the performance of the system under the effect of several important factors including sampling rate and window size on several Android smart phones. The tests are performed on different subjects for activities of walking, running, sitting, standing and biking. We evaluated the performance of the activity recognition system using the Naive Bayes classifier, and next we utilized Clustered KNN and Decision Tree algorithms. According to the results, Naïve Bayes provides not satisfactory results whereas Clustered KNN gives promising results compared to the previous studies and even with the ones which consider offline classification. Additionally, Decision Tree results are also comparable with the results of Clustered KNN.