Mining top high utility association rules using Binary Differential Evolution

Abstract

In this paper, in a first-of-its-kind-study a Binary Differential Evolution (BDE) and an Adaptive Binary Differential Evolution (ABDE)-based top non-redundant high utility association rule mining (TNR-HUARM) algorithms are proposed. A real-life OnlineRetail dataset is analyzed as an application of TNR-HUARM for customer segmentation based on value, i.e. utility. Apart from this dataset, also tested the effectiveness of our proposed models on six high utility benchmark datasets. The results show that BDE based TNR-HUARM numerically outperformed ABDE based TNR-HUARM on all seven datasets. But, when a t-test is performed, it turned out that the null hypothesis is not rejected. Therefore, Differential Evolution and Adaptive Differential Evolution based rule miners are found to be statistically equal in six out of seven datasets. The proposed algorithms were compared against two state-of-the-art algorithms and mixed results were obtained.