AN OPTIMISED MACHINE LEARNING FRAMEWORK FOR CREDIT CARD FRAUD DETECTION USING HYBRID ENSEMBLE TECHNIQUES AND REAL-WORLD DATA ENGINEERING

The rapid growth of digital payment platforms has been accompanied by a surge in sophisticated credit card fraud, posing significant risks to the financial sector. Conventional fraud detection mechanisms often fall short due to challenges such as severe class imbalance, constantly evolving fraudulent tactics, and the pressing need for real-time responses. This study presents an optimised machine learning framework that integrates ensemble classifiers such as Random Forest, XGBoost, Support Vector Machine, and Logistic Regression, combined with data-centric techniques including SMOTE (Synthetic Minority Oversampling Technique), Recursive Feature Elimination (RFE), and Z-score normalisation. The framework, designed with these challenges in mind has proven to be highly effective. Experimental results demonstrate the improved predictive performance, of the proposed hybrid ensemble, achieving an F1-score of 0.93 and a ROC-AUC of 0.98. In addition, SHAP (SHapley Additive exPlanations) analysis enhances interpretability by providing transparent, model-agnostic insights into decision-making. The framework therefore, outperforms traditional baselines and provides a scalable solution suitable for real-time credit card fraud detection in modern financial systems