A novel pruning approach using expert knowledge for data-specific pruning

摘要: Classification is an important data mining task that discovers hidden knowledge from the labeled datasets. Most approaches to pruning assume all dataset are equally uniform and important, so they apply equal However, in real-world classification problems, datasets not considering rate during tends generate a decision tree with large size high misclassification rate. We approach problem by first investigating properties of each then deriving data-specific value using expert which used design techniques prune trees close perfection. An efficient algorithm dubbed EKBP proposed very general as we free use any learning base classifier. have implemented our solution experimentally verified its effectiveness forty real world benchmark UCI machine repository. In these experiments, shows it can dramatically reduce while enhancing or retaining level accuracy.