A new decision tree pre-pruning method based on nodes probabilities

Abstract

One of the most well-known and effective data mining approaches is the decision tree. Many researchers have established and thoroughly investigated this technique. On the other hand, some decision tree algorithms may yield a complex structure that is difficult to comprehend. In addition, data misclassification is common during the learning process. Pruning can be utilized as a fundamental procedure to solve this problem. To improve generalization, it eliminates the use of noisy, contradictory data. In this paper, we propose a new pre-pruning method that prunes weak nodes with a high probability. The experimental results are verified using 24 benchmark datasets from the UCI machine learning repository. The results indicate that our new tree pruning method is a feasible way of pruning decision trees.