Research and Improvement of Decision Tree’s Prune Algorithm

This paper studies the exist prune algorithms of decision tree in machine learning of artificial intelligence, and proposes a new prune algorithm—failure-node prune algorithm. Usually, when a decision tree is pruned, the number of the tree’s nodes will decrease and the correctness will also be cut down. The failure-node prune algorithm proposed in this paper can increase the correctness while the number of the tree’s nodes is decreased when the data sets are quite deficient. Four typical databases obtained from CMU are used to do the experiments on the ID3 algorithm, the expected-error prune algorithm and the failure-node prune algorithm. In the experiments, it divides a data set into three parts, used two of them to form the training set and the third part as the testing set. The results of the experiments suggest the failure-node prune algorithm has a quite good pruning effect.