A Scalable RF-XGBoost Framework for Financial Fraud Mitigation

Machine learning offers the ability to analyze large financial data and extract hidden knowledge for anomalous patterns. In the current study, we propose an ensemble machine learning framework using eXtreme Gradient Boosting (XGBoost) and random forest for the detection of financial frauds. The model used an adaptive synthetic (ADASYN) algorithm over sampling technique with out-of-bag scoring on a grid search algorithm for the 11 input features. We tested our framework with IEEE-CIS fraud detection benchmark dataset from Kaggle. The performance metrics were F1-Score = 0.9982, area under curve = 0.9999, recall = 1.0, precision = 0.9965, and accuracy = 0.9999. A detailed comparison with state-of-the-art techniques, such as decision tree, logistic regression, gradient boosting, and particle swarm optimization models, shows scalability and robustness of the framework.