An Explainable Ensemble Model for Credit Card Fraud Detection Using Advanced Data Balancing and Feature Selection Technique

The rise of eCommerce and digital banking has transformed financial transactions, making online payments an essential part of modern economies. However, the increasing reliance on credit cards has led to a surge in fraudulent activities, highlighting the critical need for efficient fraud detection to secure digital payment systems. Traditional methods often struggle with accuracy, interpretability, and adaptability due to data imbalance and high-dimensional feature spaces. To address these challenges, this study introduces an advanced machine learning pipeline that integrates ensemble learning with optimized data balancing and feature selection techniques. Nine machine learning models were trained, and the top three- Random Forest, Gradient Boosting Machine (GBM), and CatBoost Classifier-were selected based on accuracy and combined into a Voting Classifier to leverage their complementary strengths. The ensemble model achieved 98.11% accuracy, 96.60% precision, 99.74% recall, and 98.14% F1-score, demonstrating its effectiveness in fraud detection. Additionally, Explainable Artificial Intelligence (XAI) techniques, LIME and SHAP, were employed to enhance model interpretability and identify key fraud-related features. This study contributes to improving fraud detection in highly imbalanced datasets, strengthening digital payment security, and increasing trust in online transactions.