DRAC: A Direct Rule Mining Approach for Associative Classification

The application of associative rule mining in classification (associative classification) has demonstrated its power in recent years. The current associative classifier building often adopts three phases: Rule Generation, Building Classifier and Classification. Unfortunately, in rule generation phase, a large number of rules are usually produced, which could not only slow down the mining process but also bring challenge to pruning and storing such magnitude of rules. In this paper, we propose the DRAC, a Direct Rules mining approach for Associative Classification, to tackle the efficiency of associative classification problem. DRAC can mine the high quality non-redundant rule set directly. At the same time, it also adopts the multiple strong class association rules to classify the unlabeled dataset correspondingly. The experimental results show that DRAC is more efficient than traditional approach CBA without losing of accuracy.

I. Introduction

Association rule mining and classification are two important data mining techniques used in amount application areas, including finical market, bioinformatics, web analysis, and so on. The goal of association rule mining is to find the rules in the database with confidence and support greater than the user specified threshold [1]. Classification is used to build a classifier by analyzing the given training datasets with a class label, and predict the unlabeled objects. In recent years, a new approach called associative classification is proposed to integrate association rule mining and classification, such as CBA[7], CMAR[8]. This new technique was found to be competitive with traditional classification methods, such as C4.5 and SVM due to its higher accuracy.