An Explainable AI and Machine Learning Framework for Intelligent Financial Decision Making, FinTech Analytics, Risk Prediction, and Automated Financial Services

This study proposes an explainable machine learning model for credit default prediction and financial decision support using UCI Credit Card Defaults data set. A total of 30,000 customers were analyzed, along with 23 predictors, and a 22.12% prevalence of default. In this leakage-resistant development-validation-locked-test (DVLT) design, 4 classifiers were evaluated and compared for selection – Logistic Regression, Random Forest, Extra Trees and HistGradientBoosting – using PR-AUC as the primary selection criterion. HistGradientBoosting was chosen as a best model by the following criteria: validation PR-AUC = 0.5492, validation ROC-AUC = 0.7744 and validation Brier score = 0.1366. On the locked test set, the calibrated model achieved PR-AUC 0.5585, ROC-AUC 0.7835, recall 0.8554, F2 0.6441, and MCC 0.2932 at a validation-derived threshold of 0.1274. The most important predictor was recent repayment status (PAY_0) according to permutation analysis. The observed default rates in the risk bands derived from validation range from 8.92% for low-risk to 70.31% for critical-risk. Hence the framework facilitates high sensitivity screening, understandable risk prioritization and human-supervised FinTech decision flow.: Explainable Artificial Intelligence, Credit Risk, Fintech Analytics, Machine Learning, Default Prediction, Decision Support