Top-down Decision Tree Pruning Method Research

Decision trees as an important branch of machine learning has been used in several areassuccessfully. By the limitation of small training set decision trees often overfitting the trainingset. This phenomenon weaken accuracy of decision trees. In order to overcome those defectsof decision tree learning, it is usually using decision trees pruning as next step of the decisiontrees learning algorithm for optimizing decision trees.Now common used decision trees pruning algorithm is based on statistical analysis. Thesmall training set is lack of statistics characteristics, and it leads pruning methods to failure.By summarized and analyzed the previous work, this paper presents a top-down decision treeincremental pruning method. This method based on incremental learning. To remain certainlyrules and remove uncertainly rules, the method comparing the different between their changerate of information gain when add new examples to training set. In addition the pruning isfrom root to leave to save pruning time.Top-down decision tree incremental pruning method is independent of statisticscharacteristics of the training set. It is a robust pruning method. The experimental resultsshow that Top-down decision tree incremental pruning method maintain well balance betweenaccuracy and size of pruned decision trees, and the pruned decision tree is better than thosetraditional methods in classification problems.