A novel approach of mining strong jumping emerging patterns based on BSC-tree

Abstract

It is a great challenge to discover strong jumping emerging patterns (SJEPs) from a high-dimensional dataset because of the huge pattern space. In this article, we propose a dynamically growing contrast pattern tree (DGCP-tree) structure to store grown patterns and their path codes arrays with 1-bit counts, which are from the constructed bit string compression tree. A method of mining SJEPs based on DGCP-tree is developed. In order to reduce the pattern search space, we introduce a novel pattern pruning method, which dramatically reduces non-minimal jumping emerging patterns (JEPs) during the mining process. Experiments are performed on three real cancer datasets and three datasets from the University of California, Irvine machine-learning repository. Compared with the well-known CP-tree method, the results show that the proposed method is substantially faster, able to handle higher-dimensional datasets and to prune more non-minimal JEPs.