Affiliations: [a] Center for Business and Information Technologies, University of Louisiana at Lafayette, P.O. Box 44932, Lafayette, LA 70504-4932, USA. E-mail: [email protected] | [b] Department of Electrical and Computer Engineering, University of Miami, 1251 Memorial Drive, Coral Gables, FL 33146, USA. E-mail: [email protected]
Correspondence:
[*]
Corresponding author.
Abstract: The input of a classical application of association mining is a large set of transactions, each consisting of a list of items a customer has paid for at a supermarket checkout desk. The goal is to identify groups of items (“itemsets”) that frequently co-occur in the same shopping carts. This paper focuses on an aspect that has so far received relatively little attention: the composition of the list of frequent itemsets may change in time as the purchasing habits get affected by fashion, season, and introduction of new products. We investigate (1) heuristics for the detection of such changes in time-ordered databases and (2) techniques that update the set of frequent itemsets when the change is detected. As the main performance criterion, we use the accuracy with which our program maintains the current list of frequent itemsets in a time-varying environment.
Keywords: association mining, change detection, similarity metrics
