Evaluating time variations to identify valuable association rules in market basket analysis

Abstract

Data mining techniques are extensively applied to retail market to discover association rules between products. Indicators such as lift and confidence have been defined to evaluate and promote valuable association rules. The aim of this paper is to propose a method of identifying the most valuable association rules among the large set of discovered association rules. To achieve the above, our study proposes new indicators, concerning a) the variability of an association rule, that expresses the average variability of both lift and confidence indicators of an association rule and b) the variability of a product, that expresses the average variability of the association rules that the specific product participates. Based on these indicators, our proposed method identifies the most inflexible association rules (rules that present low degree of variability). Our method examines these inflexible rules, and based also on their confidence and lift indicators, provides the most valuable association rules regarding consumer behavior. The above method is applied to annual customer transaction data that have been collected by the use of loyalty cards from a known Greek supermarket chain. To discover the association rules, the apriori algorithm is used. In the paper, the most valuable rules that are identified by the proposed method are compared to those identified by the classical approach that uses only the confidence and lift indicators. The comparison exhibits the advantages of our proposed method.