Integrating classification capability and reliability in associative classification: A beta-stronger model

Mining class association rules is an important task for associative classification and plays a key role in rule-based decision support systems. Most of the existing methods try the best to mine rules with high reliability but ignore their capability for classifying potential objects. This paper defines a concept of -stronger relationship, and proposes a new method that integrates classification capability and classification reliability in rule discovery. The method takes advantage of rough classification method to generate frequent items and rules, and calculate their support and confidence degrees. We propose two new theorems to prune redundant frequent items and a concept of indiscernibility relationship between rules to prune redundant rules. The pruning theorems afford the associative classifier with good classification capability. The experiment shows that the proposed method generates a smaller frequent item set and significantly enhances the classification performance.