Abstract
This paper investigates the performance of ensemble machine learning algorithms - Random Forest, Gradient Boosting, and XGBoost - for solving the dual-classification problem of credit scoring and fraud detection within financial decision support systems. The study utilizes an event-driven architecture for real-time data processing, evaluated on retail lending application data through cross-validation and hyperparameter optimization. Experimental results demonstrate that Random Forest and XGBoost achieve high predictive accuracy for credit scoring, while boosting models effectively handle fraud detection. Furthermore, cost-sensitive threshold calibration optimizes risk management, and feature importance analysis reveals that behavioral and data consistency indicators outweigh traditional credit scores. The findings offer a robust methodological and architectural contribution to real-time financial decision support systems, enhancing automated risk assessment and fraud mitigation in digital lending.