Improved Market Basket Analysis with Utility Mining

Abstract

Data Mining is a tool for retrieving novel and useful information contained in huge data stores. Conventional approaches to data mining techniques have majorly targeted the discovery of correlations among items that occur often in transactional databases. This method referred to as frequent item set mining believes that recurring item sets must be more significant to the user. Alternatively, in this paper, we attempt to simulate an algorithm for a recent development called Utility Mining, which studies the usefulness or utility of item sets, in addition to their frequency. This High Utility Item Set Mining facilitates the recognition of item sets having utility value greater than a lower limit specified by the seller. Though high confidence is attained, yet the limitation of our approach lies in low support. Consequentially, most useful frequent item sets might be ignored, leading to incorrect analysis of consumer behavior pattern. In the future, we intend to devise an ensemble algorithm which can tackle this issue.