THE COMPARATIVE ANALYSIS OF DECISION TREES AND NEURAL NETWORKS METHODS IN THE CREDIT INSTITUTIONS CLASSIFICATION PROBLEM

Abstract

In recent years, decision trees and neural network have been widely used in computer vision problems such as object recognition, text classification, gesture recognition, spam detection, semantic segmentation and data clustering. The article discusses the decision tree and neural networks methods application in the problem of classifying credit institutions as economic security objects. The analysis results of data on credit institutions activities of using different methods of decision trees: C5, CHAID, C&R and QUEST, as well as neural networks are presented. The highest overall classification accuracy of the analyzed objects was achieved using the C5 decision tree algorithm and amounted to 81 %. The overall classification accuracy using the CHAID algorithm was 68 %, the C&R algorithm was 71 %, and the QUEST algorithm was 66 %. Based on the C5 algorithm results, a set of rules was generated to determine whether a bank belongs to a certain class. According to the methods of decision trees and neural networks, the most informative performance indicators of credit institutions were selected in terms of their division into two classes: trustworthy and high-risk.