Author(s): Tshub V. S.

Rubric: Information technology

DOI: 10.21777/2500-2112-2023-3-81-92

Release: 2023-3 (44)

Pages: 81-92

Keywords: artificial intelligence, machine learning, dropout, short-term memory, correlation matrix, credit risks

Annotation: The article describes the comparative effectiveness of automated credit risk assessment by modern machine learning methods (Naive Bayes classifier, k-nearest neighbors, logistic regression, “random forest” and deep neural network). A database was modeled from available open sources, samples were made, and data was preprocessed. Machine learning models have been trained and tested, and compared by quality metrics. Based on the test results, the parameters of the machine learning model were optimized by a complete search of combinations of statistical models (rule-based method, k-nearest neighbor method, logistic regression, discriminant analysis, naive Bayes classifier, neural networks and decision trees). Each of the traditional classifiers (naive Bayes classifier, k-nearest neighbor method, logistic regression, “random forest”, deep neural network) was improved with the use of standardization mechanisms and dimensionality reduction by the principal component method. Based on the results of a comparative analysis of machine learning models, the best result was demonstrated by a neural network model with optimized parameters, which provided the best indicators for all evaluation metrics.

Bibliography: Tshub V..S.. COMPARATIVE ANALYSIS OF MACHINE LEARNING METHODS IN CREDIT RISK ASSESSMENT // Education Resources and Technologies. – 2023. – № 3 (44). – С. 81-92. doi: 10.21777/2500-2112-2023-3-81-92

Text article and list references