Information Bottleneck Based Trajectory Clustering And Visualization

Trajectory clustering aims to assign the similar trajectories into the same cluster,which is a very important technique for trajectory analysis.In fact,most of the clustering algorithms face with two challenging issues that are selecting an appropriate trajectory distance measure and determining an optimal number of the clusters.In this paper,we propose to utilize the agglomerative Information Bottleneck(aIB)principle for trajectory clustering,which can obtain an optimal number of clusters without a direct use of distance measure.By introducing a feature space of the trajectory dataset,aIB builds a “bottleneck”by maximizing the mutual information between clustering result and feature space and,minimizing the mutual information between original trajectory dataset and clustering result to successfully reach the goal.We propose to use a statistically representation of the trajectory shape to build the information channel among the original trajectory dataset,feature space and the clustering result by utilizing kernel density estimation(KDE),and then to perform the aIB based trajectory clustering.In addition,we propose an extension of aIB to cope with the clustering on trajectories with outliers,namely an extension of agglomerative Information Bottleneck based trajectory clustering with outlier,XaIBO.Extensive experiments on synthetic,simulated and real trajectory dataset have shown that XaIBO performs better than aIB when the trajectory dataset has outliers,and than the clustering approaches widely used in practice.Undoubtedly,visualization of the trajectory clustering outputs is very important and some researches have been carried out on visualization of the clustering results.Still importantly,the research on visualizing the procedure of clustering,which is also of great value,is little touched.In this paper,considering XaIBO is based on the abstract information theory tool,we propose to design a novel procedure visualization tool,which can comprehensively illustrate the clustering scheme of XaIBO,to help users understand the clustering approach vividly and clearly.The point of the proposed metaphor makes use of the visualization,together with rich interactions,to demonstrate the iterative clustering procedure,the corresponding results and the clustering results.The experiment base on real trajectory data demonstrates the effectiveness of our visualization tool for trajectory analysis.