Mining Frequent 3D Sequential Patterns

We propose a mining approach, MSP, to find the maximal sequential 3D patterns with the constraints of minimum support and minimum confidence. Each pattern is a group of similar sequential 3D objects appearing in a given dataset. Mining sequential patterns in terms of 3D coordinates is important and meaningful in many real-life applications. MSP finds out the maximal patterns in terms of both length and frequency without loss. MSP involves three stages: generating seeds with pairwise pattern mining, vertical extension to detect all hits with a depth-first search and horizontal extension to extend the pattern length without loss of hits. Furthermore, we propose a method to automatically detect proper settings in order to adapt MSP to various datasets. The experiments on protein chains and synthetic data show MSP significantly outperforms the alternative methods. We apply MSP to protein family classification and pattern mining in spatial moving objects. The obtained patterns correctly classify the protein families on all the tested binary-class datasets. Sample patterns in protein structures and spatial moving objects are presented