Improvement of Apriori Algorithm for Missing Itemset Identification and Faster Execution

摘要

Association rule mining (ARM) is an important data mining strategy to analyze the relationship among the items. Apriori algorithm is the most used approach to implement association rule mining. We identified two major issues of Apriori. Apriori follows an iterative approach consisting of multiple database scans for searching frequent itemsets that satisfy certain threshold criteria. The same predefined threshold value is maintained throughout the repetitive stages of the Apriori method and hence there is a huge possibility of discarding higher-order itemsets, though all of its sub-itemsets are frequent. Some of these ignored itemsets if used intelligently have a huge potential for business value addition. Furthermore, in the Apriori procedure, an exponential number of computations is required to check whether an item is important or not and that makes the entire pattern mining system costly. In this study first, we identify the hidden business-critical item sets that are otherwise ignored in the traditional Apriori process. Furthermore, a novel approach is proposed here to evaluate whether an item is interesting or not at a considerably reduced computational time.