Modern Approaches to Identifying Suspicious Companies in Corporate Client Lending Using Machine Learning Methods

Digital technologies are the driver of the development of any business in modern conditions, including the sphere of lending to corporate clients. On the one hand, automation can significantly reduce the time needed to make decisions on loans, thereby increasing customer loyalty, but on the other hand, new, more innovative types of fraud are emerging in the digital environment. The relevance of the study is due to the fact that in the context of the active development of the digital credit process, the issue of creating and implementing predictive models to identify unscrupulous borrowers of banks at the stage of making a decision on granting a loan, as well as in the process of monitoring their activities, remains acute. One of the solutions to the above problems is the implementation of an anti-fraud model using machine learning methods. The object of the study is the bank's relations with corporate clients who are potential or current borrowers. The subject of the study is empirical data on the identified abuses and the experience of taking preventive measures in the field of lending to corporate clients of banks. The purpose of this work is to study modern approaches to identify dubious companies in the field of lending using machine learning methods. Tasks of the Master's thesis: • research of theoretical aspects of problems related to the tasks of identifying dubious companies in the field of lending; • analysis of the types of violations and their characteristics allowed in the field of corporate customer lending; • identification and classification of signs that determine the affiliation of corporate clients to dubious companies based on the analysis of cases from the practice of banking institutions; • analysis of machine learning methods used to build a model for identifying dubious companies in the field of corporate customer lending; • consideration of the applied aspects of the implementation of the predictive anti-fraud model for the identification of dubious companies. The methodological basis of the research consists of the following methods: the transition from the abstract to the concrete, analysis and evaluation of empirical data (case method), comparison, classification, modeling related to general scientific, as well as a number of private scientific methods, including the OSINT method and the method of statistical information processing used in analyzing the level of losses from fraudulent activities in the banking sector Russia. In the first chapter of the work, the theoretical aspects of the problems associated with the tasks of identifying dubious companies in the field of lending are considered, concepts and operational definitions are given. In the second chapter, an analysis of the types of violations and their characteristics allowed in the field of lending to corporate clients is carried out, as well as the features of the implementation of the digital credit process in banking institutions are considered. In the main part of the work, using the case method, examples from the practical activities of the teaching staff of the HSE Institute of Security Problems are analyzed, and on their basis, signs are identified that allow companies to be classified as questionable. The applied aspects of the implementation of a predictive anti-fraud model for identifying dubious companies using machine learning methods are also considered. The applied significance of the research results lies in the possibilities of their use in practice in the implementation of the digital credit process in commercial banks. The work contains an introduction, three chapters, a conclusion, and a list of used literature, including 34 sources. The volume of work is only 74 pages of typewritten text (56 pages the main text), including 15 figures, 7 tables and 3 appendices.