Perbandingan Algoritme Pruning pada Decision Tree yang Dikembangkan dengan Algoritme CART

Abstract

Pruning is part of the development of decision tree. As decision tree is developed, some nodes became outliers as the results of noise data. Implementation of the decision tree pruning, can reduce the noise and outliers on the initial decision tree so that it can improve the accuracy of the data classification. Therefore the selection of proper pruning algorithm needs to be done to get the maximum results of the classification. This experiment uses data from the company's customer credit providers. The data obtained from the data bank at the University of California. Data used in this experiment has twenty variables with two classes and 1000 instances. The data contain thirteen qualitative variables and the rest is a numeric data. The data is a good for use because it does not have a missing value. The experiment compared three pruning algorithm, Cost Complexity Pruning (CCP), Reduced Error Pruning (REP), Error Based Pruning (EBP). Those pruning algorithms do prune to the decision tree that was developed with the Classification and Regression Tree (CART) algorithm. The comparison of those algorithms is done repeatedly on the data with different conditions both in terms of the instance number and the data variables. Comparison of the algorithm includes a comparison of the accuracy of the decision tree, and the process time of pruning algorithm. The experiment’s result shows the average error rate of that the REP algorithm will produce the smallest error rate. Although the error rate of REP algorithm is the smallest, the difference value between ERP’s and EBP’s error rate is only 0.5%. Even though they have almost similar error rate, EBP algorithm proposes more simple decision tree than REP algorithm does. Keyword : Decision tree, Classification and Regression Tree (CART), Cost Complexity Pruning (CCP), Reduced Error Pruning (REP), Error Based Pruning (EBP).