MOPNAR-II: An Improved Multi-Objective Evolutionary Algorithm for Mining Positive and Negative Association Rules

Abstract

This paper designs a multi-objective evolutionary algorithm called MOPNAR-II to mine positive and negative association rules. The main idea of MOPNAR-II is to improve the related operations of multi-objective optimization approach using reference-point based non-dominated sorting approach to guarantee the diversity of rules obtained. In MOPNAR-II, a gene-oriented crossover strategy is utilized to perform extended crossover operation between two intervals, so as to improve the ability of exploitation, while the mutation of rules and the limitation of boundary are also improved. In the experiments, MOPNAR-II is evaluated over two algorithms that are for mining positive and negative association rules and a classical algorithm Apriori on real-world datasets. Results show that MOPNAR-II performs better in terms of quality measures, which prove the validity of our algorithm.