Group-based approach for class prediction in associative classification

Classification using association is a data mining approach that integrates association rule discovery and classification. Associative classification includes three main steps; rule discovery, rule pruning and class prediction. This paper addresses the class prediction by introducing Joint Confidence Support Class Prediction (JCSCP) method that splits generated rules into groups based on the class label. The identified groups are given weights to indicate their importance to the classification. Experiments on UCI and Reuters datasets show that the proposed method outperformed the use of single rule method in producing accurate class prediction.