INDIVIDUAL’S CREDITWORTHINESS ASSESSING MODELS CONSTRUCTION AND STUDY BASED ON BOOTSTRAPING METHOD

Abstract

This article solves the problem of constructing and studying models for individual’s creditworthiness assessing. The relevance of solving this problem on the intelligent modeling technologies basis: neural networks, decision trees, logistic regression is noted. The initial data for the models constructing was a set of 35 columns and 149,000 rows. The model’s construction and study were carried out in the Deductor Analytical Platform. Each model was tested on data set of 54827 records. For each model we constructed the corresponding classification matrices and calculated the 1st, 2nd kind errors, and the general error of the models. In terms of minimizing these errors, logistic regression showed the worst results, and the neural network showed the best. In addition, the constructed models’ effectiveness was evaluated according to the «Income from loans» criterion. According to this criterion, the neural network model was also the best. Thus, the study results showed that to maximize profits and minimize classification errors, it is advisable to use a neural network model. This indicates its effectiveness and practical use possibility in intelligent decision-making support systems for assessing the potential borrowers’ creditworthiness.