CorClass: Correlated Association Rule Mining for Classification

Abstract

AbstractA novel algorithm, CorClass, that integrates association rule mining with classification, is presented. It first discovers all correlated association rules (adapting a technique by Morishita and Sese) and then applies the discovered rule sets to classify unseen data. The key advantage of CorClass, as compared to other techniques for associative classification, is that CorClass directly finds the associations rules for classification by employing a branch-and-bound algorithm. Previous techniques (such as CBA [1] and CMAR [2]) first discover all association rules satisfying a minimum support and confidence threshold and then post-process them to retain the best rules.CorClass is experimentally evaluated and compared to existing associative classification algorithms such as CBA [1], CMAR [2] and rule induction algorithms such as Ripper [3], PART [4] and C4.5 [5].KeywordsAssociation RuleCorrelation MeasureCombination StrategyRule DiscoveryDecision ListThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.