Ensemble Methods for Predicting Continuous Cash Flow in Credit Risk Assessment

Accurate prediction of continuous cash flow is a pivotal component in credit risk assessment, directly informing lending decisions and risk management. This study evaluates the predictive efficacy of ensemble learning algorithms—Random Forest and XGBoost (latter implemented as HistGradientBoostingRegressor in scikit-learn) —against traditional baseline models, namely Linear Regression and Decision Tree Regressor, in forecasting cumulative monthly loan repayments over a 12-month horizon. An ETL pipeline with categorical encoding is employed alongside rigorous data preparation steps addressing missingness and multicollinearity via Variance Inflation Factor analysis. Model hyperparameters are optimized through scikit-learn’s GridSearchCV. Results demonstrate the superior accuracy and robustness of ensemble methods, emphasizing their applicability in practical credit risk frameworks.