Finding Contrast Patterns in Imbalanced Classification based on Sliding Window

Abstract

In the process of the contrast patterns mining, people usually assume that the datasets distribu- tion is basic balance, but in the real world, there are many data sets which class distribution is imbalanced. Considering the problem of contrast patterns mining on the imbalanced data sets, in this paper, we introduce the balance factor, give a new defined contrast patterns called balance emerging patterns(BEPs for short) which suitable for the imbalanced data sets, and propose a new algorithm WBEPM, it construct a sliding win- dow to mine the BEPs on the imbalanced datasets. Experimental results show that the proposed algorithm has a better mining effect than the algorithm for original simple contrast patterns mining, the classification accu- racy of the BEPs classifier is higher than that of the previous contrast patterns classifier when deal with the imbalanced data sets. KEYWORD: Class Imbalance; Balance Factor; Sliding Window; Contrast Pattern Tree