Interesting Pattern Mining Using Item Influence

Abstract

Interesting patterns are very much needed in mining of significant association rules which play a big role in knowledge discovery. Frequency based pattern mining techniques such as support often lead to the generation of huge number of patterns including the uninteresting ones with high dissociation. Though dissociation is used to distinguish between two patterns having equal support, but if both of support and dissociation are same then it becomes very difficult to distinguish them. To overcome these types of problem we have introduced a new method of pattern mining based on the concept of item influence. How many other distinct items have been appeared with an item throughout the dataset is referred to its item influence (ii). The proposed method includes three consecutive steps such as- (1) measurement of item influence for all of the items present in the database, (2) calculation of transaction influence (ti) for all of the transactions present in the database using item influence and (3) measurement of influential weights (iw) for all of the generated itemsets. Pruning is done based on the minimum threshold value corresponding to influential weights. Experimental analysis shows the effectiveness of the method.