Mining association rules directly using ACO without generating frequent itemsets

Abstract

Association rule mining is one of the significant tasks in data mining. In literature, several approaches for finding interesting association rules have been proposed. Finding association rules is a two phase process. The first phase finds frequent itemsets or patterns and the second phase generates association rules. The phase that detects the frequent itemsets consumes more time and efforts. Thus performance and efficiency of an approach for generating association rules depends upon the efficiency of the approach used to find frequent itemsets in the first phase. The present paper proposes an approach that generates association rules directly without undergoing through this two phase process. ACO based methodology is applied to generate association rules directly. Item database is converted into a directed graph and then ACO is applied to generate association rules in a single step without generating large number of candidate itemsets. The algorithm is inspired by the AntMiner approach used for generating classification rules.