Association rule mining using multi-objective evolutionary algorithms: Strengths and challenges

Abstract

Association rule mining based on support and confidence generates a large number of rules. However, post analysis is required to obtain interesting rules as many of the generated rules are useless. We pose mining association rules as multi-objective optimization problem where objective functions are rule interestingness measures and use NSGA-II, a well known multi-objective evolutionary algorithm (MOEA), to solve the problem. We compare our results vis-à-vis results obtained by a traditional rule mining algorithm - Apriori and contrary to the other works reported in the literature clearly highlight the quality of obtained rules and challenges while using MOEAs for mining association rules. Though none of the algorithm emerged as clear winner, some of the rules obtained by MOEA could not be obtained by traditional data mining algorithm. We treat the whole process from data mining perspective and discuss the pitfalls responsible for relatively poor performance of the MOEA which has been shown as a good performer in other paradigms.