An Explainable and Robust Loan Approval Prediction System using Hybrid XGBoost and Heuristic Risk Assessment

Loan approval is a key financial process that affects both banks and people applying for loans. Old ways of checking loans depend a lot on manual checks and strict rules, which can lead to slow processes, unfair decisions, and expensive operations. This paper introduces a new system that uses data and smart technology to predict loan approvals. The system uses Extreme Gradient Boosting, or XGBoost, along with a combination of decision tools. The system has three main parts: (1)automatically checking documents with OCR to confirmwhat applicants say, (2) using a calibrated XGBoost model toestimate the chance of risk, and (3) applying penalty rules basedon financial numbers. Testing this system on real loan datashows it can predict approvals with 98% accuracy. It alsoimproves how well risk is measured because of the calibration.Compared to traditional methods like Random Forest andLogistic Regression, this new approach is better at makingaccurate predictions and keeping operations safe.