Significant Association Rule Mining with High Associability

Traditional support-confidence framework based association rule mining approaches often generate huge number of rules including the insignificant ones. These insignificant association rules are useless in knowledge discovery. One of the main reasons behind the generation of insignificant association rules is the use of confidence as the lonely interestingness measure in defining rule significance. Confidence only expresses the strength of the probability between the antecedent and consequent itemsets andignores their associability. Associability refers to closeness between the antecedent and consequent itemsets of an association rule. Assessment of associability against dissociation is an effective technique in this respect. Traditional approaches follow rule pruning with fixed maximum dissociation threshold. In reality, association rules formed with large antecedent or consequent possess high dissociation by nature. The fact leads to lose of valuable association rules. To solve the problems, this paper introduces the concept of flexible dissociation and thereafter a novel framework of significant association rule mining with high associability. Rule pruning with flexible dissociation and confidence ensures high associability of the rules. Experimental studies on real world datasets also show the effectiveness of the proposed approach.