Fast algorithms for mining Strong Jumping Emerging Patterns using the contrast pattern tree

Abstract

Efficient mining of Strong Jumping Emerging Patterns (SJEPs) is useful for constructing accurate classifiers. The method for mining SJEPs based on a contrast pattern tree structure (CP-Tree) has been demonstrated to perform extremely well for a low-dimensional dataset. In the method, a large number of non-minimal JEPs are generated during the mining process. So, it is unable to handle higher-dimensional attributes. In this paper, we propose a novel pattern pruning technique that dramatically reduces the search space. The CP-tree method is greatly improved by the proposed pattern pruning technique. Experiments are performed on two high-dimensional cancer datasets. Compared with the original CP-tree algorithm, the results show that the improved CP-tree algorithm is substantially faster, and able to handle higher-dimensional attributes.