Efficient statistical pruning of association rules

Abstract

Association mining is the comprehensive identification of frequent patterns in discrete tabular data. The result of association mining can be a listing of hundreds to millions of patterns, of which few are likely of interest. In this paper we present a probabilistic metric to filter association rules that can help highlight the important structure in the data. The proposed filtering technique can be combined with maximal association mining algorithms or heuristic association mining algorithms to more efficiently search for interesting association rules with lower support.