| Machine Learning Strategies for Securing Financial Transactions against Risks | |
|
MSRDG International Journal of Computer Scientific Technology & Electronics Engineering
© 2026 by MSRDG IJCSTEE Journal Volume 2 Issue 1
Year of Publication: 2026 |
Paper Download Article ID MSRDG-IJCSTEE-V2I1P105 |
|
Abstract: Financial institutions worldwide face escalating threats from sophisticated fraudsters who continuously adapt their methods to circumvent conventional rule-based detection systems. This paper introduces a multi-layer machine learning framework designed to identify and neutralise fraudulent financial transactions with high precision and minimal false-alarm rates. Our approach integrates an ensemble of Random Forest, Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM) networks, and Isolation Forest algorithms, unified through a soft-voting fusion mechanism. We address class imbalance — a pervasive challenge in fraud datasets — using a hybrid Synthetic Minority Over-sampling Technique (SMOTE) combined with edited nearest-neighbour undersampling. Feature engineering incorporates temporal transaction patterns, geospatial anomaly indicators, device-fingerprinting scores, and merchant-level risk ratings derived from historical behavioural analytics. Extensive experiments conducted on two publicly available datasets (IEEE-CIS Fraud Detection and PaySim) and one proprietary bank dataset (n = 1.2 million records) demonstrate that the proposed ensemble achieves an accuracy of 98.7%, an AUC-ROC of 0.994, a precision of 98.1%, recall of 97.9%, and an F1-score of 98.0%, outperforming six baseline models. Ablation studies confirm the complementary contribution of each sub-model. An explainability layer using SHAP (SHapley Additive exPlanations) values renders the model decisions interpretable for compliance officers. The framework is deployable in real-time streaming environments with a mean inference latency of 12.4 ms per transaction, satisfying the latency constraints of production payment systems. |
|
| Keywords: Fraud detection , Machine learning , XGBoost , LSTM , Ensemble methods, Class imbalance, SMOTE, Financial risk management, SHAP interpretability, Real-time inference | |
