Correlated High Utility Itemset Mining Based on Item Decomposition

Abstract

Correlated High utility itemset mining discovers itemsets whose correlation and utility are above the minimum thresholds. However, performance limitations relevant to memory and time result in scalability issues when exploring the search space in a large dataset. Additionally, the original database must be scanned multiple times. This paper addresses the combinatorial explosion in the search space by clustering the items and applying the mining process to each cluster separately or simultaneously. The proposed clustering-based technique discovers potential correlated items by studying the similarity between items' behaviors, which are made up of a set of utility-oriented sequential transactions carried out in a specific order. The items are first clustered using the hierarchical clustering technique, where correlated items are grouped. Then, the correlated and profitable patterns are derived by applying a pattern mining algorithm to each cluster. Experimental results show that the item-clustering technique has more efficiency and effectiveness than the transaction-clustering technique. Moreover, the proposed item-clustering algorithm outperforms the state-of-the-art CoHUIM algorithms in terms of memory and time, and the ratio of the accuracy reaches up to 99% for all cases.