Routine Based Analysis for User Classification and Location Prediction

In this paper, we propose a more practical method for classifying users and predicting their locations based upon data generated from their daily routines. Instead of calculating the similarity between users, we propose a Routine Based Classification (RBC) method to calculate the similarity between routines, and use this value to find the neighbouring routines that help to classify users. When making predictions, we do not use the Bayesian or HMM methods, which focus only on location data. Rather, we propose a State Based Prediction (SBP) method. This method emphasized user state, which is calculated based upon routine similarity and then adopted as a standard for location prediction. Experiments with the Reality Mining dataset collected by MIT show our methods use less information to achieve more accurate results.