Machine Learning-Driven Lending Decisions in Bank Consumer Finance

Abstract

This paper investigates the bank lending decision process for internet consumer finance using machine learning. It focuses on microloans and compares Logistic Regression and GBDT models for credit risk assessment. Variables are filtered and recoded via Information Value and WoE methods to enhance discrimination between defaulting and performing users. Experimental results utilizing these models predict credit risk and optimize using AUC values. Additionally, it develops a fixed-effect regression model to explore how bank-specific factors affect systemic risk, revealing that larger banks reduce risk, while higher returns, non-performing loans, and equity volatility elevate it, with inconclusive effects from leverage ratio.