Classification of Mobile Users' Activities Based on Group Similarity Perception Networks

Classifying human activities by using the data collected from a mobile phone accelerometer plays a fundamental role in a lot of applications. However, the classification effect based on a universal classifier is poor due to population diversity for numerous mobile users. To solve this problem, a classification method based on group similarity perception networks (GSPNs) is proposed. It constructs a dynamic group perception network (GPN) and takes similarity as the matching rule to choose the best classifier suitable for a given mobile user in the network. Moreover, the GSPN is dynamically optimized according to a network modification strategy in this method. The experimental results show that, compared with traditional methods, it can effectively overcome the poor classification effect of universal classifiers on mobile users arising because of the population diversity thereof.