Predicting Personal Transitional Location Based on Modified-SVM

摘要

Location prediction is an attractive research area because of its potential application in location-based services such as location-based advertisements, route planning and so on. In this paper, a method based on mining history trajectory logs is proposed to predict people's next location for a transition from current location. The training samples of this model considers temporal features containing daily and hourly features as well as spatial features like the current location which are all extracted from the historical trajectory dataset. In the learning process, the location-prediction problem can be treated as a multi-class classification and a modified support vector machine (SVM) approach is adopted to learn the classifier models. The method is tested on the Geolife project dataset which is a real life dataset of 182 users over five years. The experimental result shows that the method can provide about a 4% improvement compared with the space time features-based recurrent neural network (STF-RNN) model and a 16% improvement over the Markov models.