Ben V. The system of models for assessing the creditworthiness of borrowers-individuals

The dissertation summarizes the theoretical positions and develops a methodological approach to modeling the creditworthiness of borrowers-individuals based on artificial intelligence tools and ensemble technologies. In accordance with the proposed methodology, the application of ensemble technologies in two stages of modeling has been implemented. At the first stage, when selecting the most significant factors for assessing the borrowers' creditworthiness, an ensemble of three probabilistic neural networks, combined by the averaging algorithm, was formed. In the second stage a system of mathematical models for classification of borrowers with the use of logistic regressions, neural networks of the perceptron type and on radial basis functions was constructed. The generalization of the results of calculations of individual models was carried out within the framework of the proposed methodological approach by creating an ensemble structure with specialization of experts. Also, when constructing ensembles, the averaging algorithm and boosting were implemented and tested. Experimental study leads to the conclusion about the feasibility of ensembles of models to classify borrowers-individuals in terms of credit risk. In this case, the highest accuracy of the assessment of borrowers demonstrates the ensemble based on speciali¬zation of experts, which were represented by neural networks of different architectures.