A Discovery Method of Attractive Rules from the Tabular Structured Data

Background

Basket analysis of receipts collected from the retail business is the origin of the discovery of both frequent patterns and association rules. The rules are usually extracted from the discovered patterns. Therefore, it is important to efficiently discover the patterns. Each receipt is defined as a transaction in the analysis. Then, each transaction is composed of some items such sales items in the retail business. Each item is regarded as either of two cases in the transaction. That is, one case shows that the item is included in the transaction and the other case shows that the item is not included. Agrawal and Srikant (1994) and Han et al. (2000) propose representative discovery methods of frequent patterns using the monotonic property of the patterns. The property shows that if the pattern grows, its evaluation criterion monotonically decreases. It is called the Apriori property. On the other hand, Morzy and Zakrzewicz (1998) and Zaki et al. (1997) propose methods that speedily discover the patterns by devising storage methods for the data. Also, Koh et al. (2005) proposes a method that discovers association rules with low support but high confidence. Here, the support and the confidence are evaluation criteria of the patterns and the rules. Yan et al. (2005) proposes a discovery method of association rules based on a genetic method. The method regards the combination of an association rule including k items and the number of items in the condition part as a chromosome. It genetically discovers better association rules by using three genetic operators: select, crossover, and mutation.