Routine Based User Classification

Sensing technologies have made the tracking of users' daily trajectories a common service, such as location based service, children/elders tracking service, etc. This information can also be analyzed to discover some interesting and meaningful information about users. In this paper, we study the Routine Based Classification (RBC) approach for classifying users into different groups. For comparing two routines, we modify the Smith-Waterman alignment algorithm to increase the accuracy of similarity calculation. Term Frequency-Inverse Document Frequency (TF/IDF) is then used for classifying users based on their routines. To further improve the accuracy of classification, we propose the "group routine pattern" concept which refers to some common routines among the users of the same group. Our numerical results show that the group routine pattern concept yields higher classification accuracy than that of the Support Vector Machines (SVM) approach as well as that of an existing approach proposed in the literature.