Enhancing the Performance of Large-scale Profitable Itemset Mining using Efficient Data Structures

The process of extracting the most frequently bought items from a transactional database is termed as frequent itemset mining. Although it provides us with an idea of the best-selling itemsets, the method fails to identify the most profitable items from the database. It is not uncommon to have minimal intersection between frequent itemsets and profitable itemsets, and the process of extracting the most profitable itemsets is termed as Greater Profitable Itemset (GPI) mining. There have been various approaches to mine GPI in which [7] proposed a two-phased algorithm to optimize regeneration of GPI when the profit value of any item changes. This constituted of keeping track of the pruned items in the first phase and using it to efficiently regenerate GPI in the second phase. This paper proposes an enhancement to the way these changes are tracked by storing the pruned itemsets according to their constituent items, unlike the earlier algorithm that stored records iteration wise. By storing the itemsets according to their constituent items, we make sure that only the required items are being retrieved. In contrast, the earlier algorithm would fetch all the items pruned in any iteration, regardless of its relevance. By fetching only relevant itemset, the proposed method would significantly bring down the computational requirements.