Abstract—Clustering is the basic technique in data mining research field. However, there are just few mobility patterns based clustering techniques which are hierarchical clustering and k-means clustering. Moreover, these two techniques suffer from the so-called “curse of dimensionality”. Hence in this paper, the spectral clustering methods and the novel power symmetric normalized spectral clustering method are proposed and these three methods are used to solve the mobility pattern based clustering problem. First, the novel similarity among mobility patterns is defined in the trajectory dataset. From this novel similarity, a similarity graph can be constructed. Finally, the three proposed clustering methods are applied to this graph. Experimental results show that the clustering results of the power symmetric normalized clustering method are more well-balanced than the clustering results of the un-normalized and symmetric normalized spectral clustering methods. Moreover, the time complexity of the power symmetric normalized clustering method is also lower than the time complexity of the two spectral clustering methods.
Index Terms—Spectral clustering, graph Laplacian, similarity matrix, mobility patterns, power method.
Linh Hoang Tran is with Thu Dau Mot University, Vietnam (e-mail: firstname.lastname@example.org).
Loc Hoang Tran is with John von Neumann Institute, Vietnam (e-mail: email@example.com).
Cite: Linh Hoang Tran and Loc Hoang Tran, "Mobility Patterns Based Clustering: A Novel Approach," International Journal of Machine Learning and Computing vol. 8, no. 4, pp. 387-393, 2018.