Study on the use of different quality measures within a multi-objective evolutionary algorithm approach for emerging pattern mining in big data environments

Abstract

Emerging pattern mining is a data mining task that extracts rules describing discriminative relationships amongst variables. These rules should be understandable for the experts. Comprehensibility of a rule is traditionally determined by several objectives, which can be calculated by different measures. In this way, multi-objective evolutionary algorithms are suitable for this task. Currently, the growing amount of data makes traditional data mining tasks unable to process them in a reasonable time. These huge amounts of data make even more interesting the extraction of rules that can easily describe the underlying phenomena of this big data. So far there is only one algorithm for emerging pattern mining developed based on multi-objective evolutionary algorithms for big data, the BD-EFEP algorithm. The influence of the selection of different quality measures as objectives in the search process is analysed in this paper. The results show that the use of the combination based on Jaccard index and false positive rate is the one with the best trade-off for descriptive induction of emerging patterns. It is recommended the use of this combination of quality measure as optimisation objectives in future multi-objective evolutionary algorithm developments for emerging pattern mining focused in big data.