Resolving rule conflicts based on Naïve Bayesian model for associative classification

The rule conflict is an important issue for associative classification due to a large set of rules. In this paper, a new approach called Associative Classification with Bayes (AC-Bayes) is proposed. To address rule conflicts, AC-Bayes has two distinguished features: (1) Associative classification is improved. (2) Naïve Bayesian model is applied in process of classification. A small set of high quality rules is generated by discovering not only frequent but mutual associated itemsets. Therefore, it will reduce the opportunities of rule conflicts. AC-Bayes also selects the best n rules to predict the class value of new instances. When rule conflicts are occurred, we collect those instances covered by these matched rules to form a new training set. Then, it scans new training set to compute the posterior probabilities of each class conditioned on the test instance. The class value that maximizes the posterior probability is assigned to the test instance. The experiments' results show that the improved associative classification decreases significantly the number of rules and AC-Bayes has better average classification accuracy in comparison with associative classification and Naïve Bayesian classification.