Machine Learning Ensemble Framework for Risk-Aware Loan Approval Decisions

In the financial lending industry, achieving an optimal balance between risk and return is a key strategic objective. Under capital constraints, institutions must carefully assess trade-offs: lending to low-risk applicants ensures capital safety and stable returns but often yields lower profits due to conservative interest rates; lending to high-risk applicants may result in higher profits via elevated interest rates, yet significantly increases the risk of default. Traditional manual loan approval processes suffer from inefficiencies, high costs, and inconsistent decision-making influenced by human bias. To address these limitations, this study proposes an automated, data-driven loan approval system leveraging machine learning models to enhance accuracy and efficiency. We construct an ensemble model integrating LightGBM, XGBoost, Transformer, and MLP to predict loan approval outcomes based on applicant information. The proposed model demonstrates strong predictive performance on the test set, indicating its potential to replace manual decision-making. This approach enables institutions to automate risk assessment, optimize fund allocation, and improve operational efficiency—ultimately reducing default risks while maximizing returns and enhancing long-term competitiveness.