Fast Agglomerative Information Bottleneck Based Trajectory Clustering And Visualization

Trajectory clustering plays an important role in trajectory analysis.In fact,most of the clustering algorithms face to two challenging proplems: selecting a proper distance measure and obtaining an optimal number of the clusters.In this paper,we propose to utilize the the agglomerative Information Bottleneck principle(aIB)to do the clustering,which works well in coping with these two problems by introducing a concept of feature space.According to a IB,an optimal number of clusters are generated via maximizing the mutual information of feature space and clustering results and simultaneously minimizing the mutual information between original dataset and clusters.Moreover,we propose to speed up aIB,name fast agglomerative Information Bottleneck(faIB),by two strategies.The fisrt strategy is to do clipping on the feature space.The second is to merge more than two clusters in one iteration,which can reduce the iterations.Extensitive experiments on synthetic,simulated and real trajectory datasets demonstrate that faIB funs aboves 10 times faster than aIB,which is meaningful in large data analysis.Undoubtedly,visualization on trajectory analysis helps a lot for understanding the clustering algorithm well.In this paper,we propose a novel visualization tool for faIB algorithm,to facilitate the master of faIB vividly and clearly.The visualization tool provides a friendly user experience for controlling the parameters and results.Experiment on a real trajectory dataset show the effectiveness of our visualization tool for trajectory analysis.