SARM — Succinct Association Rule Mining: An Approach to Enhance Association Mining

Abstract

The performance of association rule mining in terms of computation time and number of redundant rules generated deteriorates as the size of database increases and/or support threshold used is smaller. In this paper, we present a new approach called SARM — succinct association rule mining, to enhance the association mining. Our approach is based on our understanding of the mining process that items become less useful as mining proceeds, and that such items can be eliminated to accelerate the mining and to reduce the number of redundant rules generated. We propose a new paradigm that an item becomes less useful when the most interesting rules involving the item have been discovered and deleting it from the mining process will not result in any significant loss of information. SARM generates a compact set of rules called succinct association rule (SAR) set that is largely free of redundant rules. SARM is efficient in association mining, especially when support threshold used is small. Experiments are conducted on both synthetic and real-life databases. SARM approach is especially suitable for applications where rules with small support may be of significant interest. We show that for such applications SAR set can be mined efficiently.

This research was supported in part by NSF Grant No. EIA-0091530, USDA RMA Grant NO. 02IE08310228, and NSF EPSCOR, Grant No. EPS-0346476.