The Formation of Model Assembly on the Basis of Specialization of Experts for Classification of Borrowers – Natural Entities

• Main Author: Ben Vladyslav P.
• Format: Article
• Language: English
• Published: Research Centre of Industrial Problems of Development of NAS of Ukraine 2018-01-01
• Series: Bìznes Inform
• Subjects: model assemblies; logit-regression; neural network
• Online Access: http://business-inform.net/export_pdf/business-inform-2018-1_0-pages-170_176.pdf

The article is aimed at researching the efficiency of the use of assembly technologies to solve the task of classifying borrowers – natural entities in relation to the level of credit risk. Features of information support of the process of solving the task of classifying borrowers – natural entities were analyzed. The necessity of application of model assemblies as one of contemporary directions of processing of big data has been substantiated. The author’s own variant of algorithm of creation of an assembly on the basis of specialization of separate models-experts has been provided. The proposed approach is implemented in two versions according to the types of models used as experts. In the first variant the assembly consists of logit-regressions, in the second variant – of neural networks. An analysis of the obtained results has proved that the described assembly structure gives an opportunity to increase the accuracy of assessment of the credit risks of borrowers, and it is expedient to use neural networks as separate models-experts.