Evaluation Index for Constructing User Preferable Decision Tree

Abstract

Decision trees are machine learning methods that are often used in data mining practice because they can construct models that are easy to interpret quickly. There is a trade-off that the size of the decision tree should be small from the viewpoint of the model's interpretability, but generally tends to be large if accuracy is emphasized. Pre and post pruning methods and their criteria and parameters are prepared as methods for adjusting the size of the decision tree. However, trial and error is required to adjust the parameters to obtain a decision tree with the preferable accuracy and size for the user. Become. In addition, it is difficult for the user to intuitively understand how well the induced decision tree can learn the training data. In order to solve this problem, we propose the evaluation index for adjusting the size and accuracy of the decision tree to be preferable for the user.