Кредитний скоринг для роздрібних позик банку

Abstract

This paper describes a credit scoring system for retail bank lending. The system predicts the probability of borrower default within twelve months. It supports loan approval, rejection, manual review, and credit limit recommendation. The solution includes a Python backend, machine learning models and the React frontend. Historical data are processed through cleaning, imputation, encoding, and scaling. Default prediction is based on logistic regression, random forest, and boosting methods. The approach improves the speed, consistency, and transparency of credit risk assessment.