Resolving Rule Conflicts with Double Induction

摘要: When applying an unordered set of classification rules, the rules may assign more than one class to a particular example. Previous methods resolving such conflicts between include using most frequent examples covered by conflicting (as done in CN2) and naive Bayes calculate probable class. An alternative way solving this problem is presented paper: generating new from rules. These newly induced are then used for classification. Experiments on number domains show that method significantly outperforms both CN2 approach Bayes.