A Pattern Tree Based Method for Mining Conditional Contrast Patterns of Multi-source Data

Li Li, Sarah M. Erfani, Christopher Leckie

Abstract: Contrast patterns are itemsets that frequently occur in one dataset while not in another. These patterns have been successfully applied to many data mining domains, such as prediction, classification and clustering. However, none of the previous studies has considered extracting contrast patterns from different types of datasets. In this paper, we introduce a new type of contrast pattern, Conditional Contrast Patterns (CCPs), which are a subset of traditional Contrast Patterns (CPs) in one kind of dataset conditioned on a property of these patterns in another kind of dataset. Accordingly, we propose an algorithm based on tree search for mining CCPs, which can compress the datasets into a tree representation. We evaluate our proposed method in comparison with two other methods (Brute force and Apriori-based methods) on a synthetic dataset as well as a real-life retail dataset. The results show that CCPs are more informative and actionable for decision makers than normal CPs, and our tree-based algorithm has the best performance in terms of efficiency.