Using SAS for Mining Indirect Associations in Data

Association rules are often used to capture significant dependencies among attributes in a variety of datasets. A typical association rule for market basket data might be of the form, "Purchasers who buy an item x will very likely buy another item y ." We have introduced a new type of dependence relationship called indirect association. A pair of items, x and y, are indirectly associated if they rarely occur together in the dataset, and there exists a set Z such that the presence of x or y in a transaction is highly dependent on the occurrence of items in Z. Indirect association has found its applications in retail, textual and stock market domains. In this paper, we describe the implementation of a prototype system for discovering indirect associations using SAS. We have tested the prototype system using masked data from Hewlett-Packard's online purchase transactions to discover interesting product purchase patterns.