ACAC: Associative Classification based on All-Confidence

Abstract

Classification and association rule mining are important data mining tasks. Associative classification integrates association rule mining and classification. Previous studies show that associative classification achieves high classification accuracy and strong flexibility. However, it often generates a huge set of rules when the minimum support is set too low. Therefore it is time consuming to select high quality rules. To deal with this problem, we propose a new associative classification approach called Associative Classification based on All-Confidence (ACAC). We use support and all-confidence to mine not only frequent but also mutual associated itemsets for classification. Therefore ACAC generates a small set of high-quality rules. Then we directly use these rules to classify new objects without pruning any rules. ACAC uses average information entropy and the number of rules to measure the combined effect of group rules. Experiment results on the Mushroom data set show that ACAC is not only efficient but also high accurate.