User centric approach to itemset utility mining in Market Basket Analysis

Business intelligence is information about a company's past performance that is used to help predict the company's future performance. It can reveal emerging trends from which the company might profit [31]. Data mining allows users to sift through the enormous amount of information available in data warehouses; it is from this sifting process that business intelligence gems may be found [31]. Within the area of data mining, the problem of deriving associations from data has received a great deal of attention. This problem is referred as “market-basket problem”. Association Rule Mining (ARM), a well-studied technique in the data mining field, identifies frequent itemsets from databases and generates association rules by assuming that all items have the same significance and frequency of occurrence in a record. However, items are actually different in many aspects in a number of real applications such as retail marketing, nutritional pattern mining, etc [26]. Rare items are less frequent items [32]. For many real world applications, however, utility of rare itemsets based on cost, profit or revenue is of importance. For extracting rare itemsets, the equal frequency based approaches like Apriori approach suffer from “rare item problem dilemma”. Utility mining aims at identifying rare itemsets with high utility. The main objective of Utility Mining is to identify the itemsets with highest utilities, by considering profit, quantity, cost or other user preferences [40]. Also valuable patterns cannot be discovered by traditional non-temporal data mining approaches that treat all the data as one large segment, with no attention paid to utilizing the time information of transactions. Now, as increasingly complex real-world problems are addressed, temporal rare itemset utility problem, are taking center stage. In many real-life applications, high-utility itemsets consist of rare items. Rare itemsets provide useful information in different decision-making domains such as business transactions, medical, security, fraudulent transactions, and retail communities. For example, in a supermarket, customers purchase microwave ovens or frying pans rarely as compared to bread, washing powder, soap. But the former transactions yield more profit for the supermarket. A retail business may be interested in identifying its most valuable customers i.e. who contribute a major fraction of overall company profit [40]. In this paper, these problems of analyzing market-basket data are considered and important contributions are presented. It is assumed that the utilities of itemsets may differ and determine the high utility itemsets based on both internal (transaction) and external utilities.