Abstract—The detection of financial fraud is one of the most serious challenges in modern financial systems and keeps on growing with increasingly complex and dynamic fraudulent activities. The paper proposes a robust hybrid model by integrating Extreme Gradient Boosting (XGBoost) and Deep Neural Networks within a stacking ensemble framework to improve the accuracy and scalability of fraud detection. It is evaluated on the pre-processed Credit Card Fraud Detection Dataset by balancing using the Synthetic Minority Over-sampling Technique (SMOTE) technique to handle the inherent class imbalance. The model achieved near-perfect performance, with a cross-validated average accuracy of 99.7% and consistently high precision, recall, and F1-scores across folds. The comparison done with previous works underlines how well the model overcomes some of the traditional challenges like data imbalance and evolving fraud patterns while scalability and adaptability for real-time applications are kept intact. Besides, the combination of advanced preprocessing techniques with ensemble learning ensures the robustness of the model in detecting fraudulent transactions. Though the model is promising, this study recognizes its limitations, such as computational complexity and dataset dependency, and proposes future research directions to optimize the model for diverse datasets and real-world constraints. Therefore, this study gives a great boost to financial fraud detection with its scalable, interpretable, and highly accurate solution and paves the path for further development in securing financial systems against fraud.
 
Keywords—financial fraud, hybrid machine learning model, Deep Neural Networks (DNNs), Extreme Gradient Boosting (XGBoost), stacking ensemble, Synthetic Minority Over-sampling Technique (SMOTE)


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).