Efficiently Mining Interesting Emerging Patterns

Abstract

Emerging patterns (EPs) are itemsets whose supports change significantly from one class to another. It has been shown that they are very powerful distinguishable features and they are very useful for constructing accurate classifiers. Previous EP mining approaches often produce a large number of EPs, which makes it very difficult to choose interesting ones manually. Usually, a post-processing filter step is applied for selecting interesting EPs based on some interestingness measures.

In this paper, we first generalize the interestingness measures for EPs, including the minimum support, the minimum growth rate, the subset relationship between EPs and the correlation based on common statistical measures such as chi-squared value. We then develop an efficient algorithm for mining only those interesting EPs, where the chi-squared test is used as heuristic to prune the search space. The experimental results show that our algorithm maintains efficiency even at low supports on data that is large, dense and has high dimensionality. They also show that the heuristic is admissible, because only unimportant EPs with low supports are ignored. Our work based on EPs for classification confirms that the discovered interesting EPs are excellent candidates for building accurate classifiers.