Pruning With Majority and Minority Properties

Classification is very important research in knowledge discovery and machine learning. The decision tree is one of the well-known data mining methods. In general, a decision tree can be grown so as to have zero error on the training data set. If there is any noise in the data set or it does not completely cover the decision space, then over-fitting occurs and the tree needs to be pruned in order to accurately generalize the test data set. In this paper, we propose a pre-pruning method with majority and minority properties for the decision tree. It uses two kinds of qualifying criteria to consider whether the ration of the highest class of a subtree is the majority of the subtree or a minority of the overall tree. New measures for these are added to the classifier with the extended data expression. Experiments show that a classifier using this pruning method can improve classification accuracy as well as reduce the size of the tree.