Semantic-Aware Clustering-based Approach of Trajectory Data Stream Mining

The rapid development of mobile sensing technologies (like GPS, RFID, accelerometer, gyroscope etc. various sensors in smart phones) has caused a rise in the large-scale capture of positioning data. These mobility data generated by heterogeneous mobile devices with embedded sensors are mostly known as trajectories. There are different algorithms to process the large amounts of mobility data for identifying mobility patterns. Even though few of these algorithms consider the use of semantic annotations on the data, none of the existing research has considered semantic annotations for online sub-trajectory clustering-based movement behavior analysis. In this paper, we incorporate semantics annotation in the raw trajectory data in order to discover various movement relationships between subtrajectories of mobile devices. We conduct experiment on a realworld data set. Along with the added advantage of semantic-aware movement behavior analysis, our method is able to identify outliers in the clustering process with almost similar performance (average recall 0.92) as classic density-based clustering algorithm DBSCAN.