Methodological aspects of the development of credit risk scoring models in the banking sector

Abstract

Introduction. The modern banking sector operates in a highly competitive environment, with increasing lending volumes and risk management requirements. In this context, assessing borrowers' creditworthiness and minimizing the risk of loan default become crucial. The development of digital technologies and the increasing volume of data necessitates the improvement of analytical tools used in credit scoring. The traditional logistic regression model, with its high level of interpretability, competes with the "black boxes" of machine learning, which offer higher prediction accuracy. This creates an urgent scientific challenge of choosing the optimal methodology for assessing credit risk.Purpose. The purpose of the study is to analyze the theoretical and methodological foundations of building scoring models based on logistic regression, as well as to identify the factors that affect the quality of credit risk assessment and the effectiveness of using these models in the banking sector.Materials and methods. The study is based on the works of authors in the field of credit scoring, risk management, and machine learning. The study uses methods of comparative analysis, generalization, and systematization of scientific approaches to building scoring models. Logistic regression is considered as the main analysis tool for predicting the probability of loan default. Modern approaches based on machine learning algorithms are also taken into account.Results. The study identified key stages in the development of scoring models, including data preparation, feature selection, and parameter estimation. It was found that logistic regression retains significant practical value due to its interpretability and compliance with regulatory requirements. The study also identified key limitations of logistic models, such as their sensitivity to data quality and limited ability to account for complex nonlinear relationships.Conclusions. Scoring models are a key tool for managing credit risk in banking. Despite the development of more complex algorithms, logistic regression remains popular due to its transparency and reliable results. Improving the efficiency of scoring models requires a comprehensive approach to data quality, proper feature selection, and model adaptation to changing environmental conditions. The core of logistic regression is the maximum likelihood method. In econometrics, the maximum likelihood function is maximized, while in machine learning, the inverse loss function is minimized. A promising area is the combined use of logistic regression with more accurate methods.