Empirically validating an adapted classification based on associations algorithm on uci data

Abstract

In recent years, extensive research has been carried out by using association rules to build more accurate classifiers. The idea behind these integrated approaches is to focus on a limited subset of association rules, i.e. those rules where the consequence of the rule is restricted to the classification class attribute. This paper aims to contribute to this integrated framework by adapting the CBA (Classification Based on Associations) algorithm. More specifically, CBA was modified by coupling it with a new measurement of the quality of association rules: i.e. intensity of implication. By means of this measurement, the sequence in which the class association rules are chosen, was changed when building the classifier. The new algorithm has been implemented and empirically tested on 16 popular datasets from the UCI Machine Learning Repository. Furthermore, the results were validated with original CBA, with C4.5 (both on original and on discretized datasets), and with Naive Bayes. The adapted CBA algorithm presented in this paper, proved to generate a lowest average error rate and produced classifiers that are more compact than original CBA.