Incremental mining class association rules using diffsets

Abstract

Class association rule (CAR) mining is to find rules that their right hand sides contain class labels. Some recent studies have proposed algorithms for mining CARs; however, they use the batch process to solve the problem. In the previous work, we proposed a method for mining CARs from incremental datasets. This method uses MECR-tree to store itemsets and rules are easy to generate based on this tree. Pre-large concept is used to reduce the number of rescan datasets and Obidsets (set of object identifiers) are used to fast compute the support of itemsets. CAR-Incre algorithm has been proposed. However, when the number of inserted records is large, this method still consumes much time to compute the intersection of Obidsets on dense datasets. In this paper, we modify CAR-Incre algorithm by using Diffsets (difference between Obidsets) instead of Obidsets. CAR-Incre is adjusted to fit Diffsets. We verify the improved algorithm by experiments.