A Novel Incremental Framework for Building Classifier Using Constraint Class Association Rules

Abstract

Associative classification (AC) performs much better than other traditional classifiers. It generates a huge number of class association rules (CARs). Since users are interested in the subset of rules, constraints are introduced in the generation of CARs. Real-world databases are record-based in which data is continuously added which demands incremental mining. Hence, constraint class association rules (CCAR) is mined from incremental data. To limit the number of rules and to remove the duplicate rules, redundant rule pruning and duplicate rule pruning techniques are applied. To improve the accuracy of the classifier, the rule selection using principality metric has been applied and the classifier is constructed with rules possessing high principality. Then, classifier is evaluated using single rule and multiple rule prediction methods and the accuracy of the proposed classifier are measured. Experimental results show that the accuracy of the proposed classifier is relatively higher when compared to other algorithms.