Modified binary cuckoo search for association rule mining

Abstract

This paper proposes a modified single-objective binary cuckoo search for association rule mining (MBCS-ARM). The proposed algorithm includes a novel representations of individuals, which tackles the problems of large dimensionality with an increasing number of attributes. The MBCS-ARM also supports the mining of rules, where intervals of attributes can either be negative or positive. It uses an objective function composed of support and confidence weighted by two parameters, which control the importance of each measure in the found rules. It is tested on eight publicly available databases, while also compared to several single-objective evolutionary algorithms, and traditional algorithms, all found in the KEEL software tool. The experiments show promising results of the MBCS-ARM, compared to other algorithms, by producing rules, which are interesting, simple, and also easy to understand, which is of great importance in domains like medicine.