Mining Efficiently Significant Classification Association Rules

Abstract

SummaryClassification Rule Mining (CRM) is a well-known Data Mining technique for the extraction of hidden Classification Rules (CRs) from a given database that is coupled with a set of pre-defined classes, the objective being to build a classifier to classify “unseen” data-records. One recent approach to CRM is to employ Association Rule Mining (ARM) techniques to identify the desired CRs, i.e. Classification Association Rule Mining (CARM). Although the advantages of accuracy and efficiency offered by CARM have been established in many papers, one major drawback is the large number of Classification Association Rules (CARs) that may be generated — up to a maximum of “2 n − n − 1” in the worst case, where n represents the number of data-attributes in a database. However, there are only a limited number, say at most k̂ in each class, of CARs that are required to distinguish between classes. The problem addressed in this chapter is how to efficiently identify the k̂ such CARs. Having a CAR list that is generated from a given database, based on the well-established “Support-Confidence” framework, a rule weighting scheme is proposed in this chapter, which assigns a score to a CAR that evaluates how significantly this CAR contributes to a single pre-defined class. Consequently a rule mining approach is presented, that addresses the above, that operates in time O(k 2 n 2) in its deterministic fashion, and O(kn) in its randomised fashion, where k represents the number of CARs in each class that are potentially significant to distinguish between classes and k ≥ k̂; as opposed to exponential time O(2 n ) — the time required in score computation to mine all k̂ CARs in a “one-by-one” manner. The experimental results show good performance regarding the accuracy of classification when using the proposed rule weighting scheme with a suggested rule ordering mechanism, and evidence that the proposed rule mining approach performs well with respect to the efficiency of computation.