Mining Correlated High-Utility Itemsets using the Cosine Measure

Abstract

High utility itemset mining (HUIM) is a problem posed to find itemsets in transaction database with high utility. However, using only utility as selection criterion makes most of the found itemsets have a very low correlation between their items, therefore it cannot be effectively applied in practice. Fast correlation high-utility itemset miner (FCHM) is an efficiency algorithm that applies correlation to HUIM problem to discover correlated high-utility itemsets (CHIs). The correlation measures used in FCHM include bond and all-confidence. This thesis proposes a new version of FCHM algorithm by using cosine measure to calculate correlation between items which is FCHM𝑐𝑜𝑠𝑖𝑛𝑒 . Experimental results on three benchmark real-life datasets show that the proposed algorithm not only significantly reduces weakly correlated itemsets but also improves running time and memory consumption.