Mining condensed rules for associative classification
Chih‐Hung Wu, Jingyi Wang, Chien-Jung Chen · 2012
This paper presents a new metric, “condenseness (cond)”, to evaluate if infrequent ruleitems that are filtered out by minsupp can also form strong ARs for classification. A new classifier, referred to as condensed association rules for classification (CARC), is developed. CARC considers the condenseness among item-sets in a ruleitem when generating ARs so that closely associated ruleitems could have chances to be discovered even they are filtered out by a higher minsupp. CARC generates ARs using a modified Apriori algorithm and develops new strategies of rule-inference. With the cond metric and strategies for rule-inference, more useful ARs can be produced and incremental trials on setting minsupp can be eliminated. Empirical evidences show that CARC mitigates the problems caused by setting too high/low minsupp and has a better performance on classification.