Система кредитного скорінгу для позик банку

Abstract

This paper describes a credit scoring system for retail loan applications in a bank. The system predicts each client’s probability of default over the next 12 months and supports loan approval or rejection decisions as well as credit limit assignment. It is implemented with a Python backend and a FastAPI-based REST API, along with a React frontend. Historical loan data are ingested from CSV files, cleaned, and transformed for modeling. We train predictive models to estimate default risk. The model outputs a default probability and score, which are used by the business logic to recommend an approval, rejection, or manual-review decision. This automation improves the speed, consistency, and transparency of credit risk assessment.