Machine Learning based Real Time Fraud Detection: Tackling Imbalanced Data with Gradient Boosting Techniques

The rise of electronic payment systems has revolutionized the financial sector but has also led to a surge in credit card fraud, imposing significant barriers for banking institutions and individuals. This study focuses on the development of real-time credit card fraud detection systems using advanced Machine Learning (ML) techniques, particularly Gradient Boosting methods such as XGBoost, LightGBM, and CatBoost. These models are wellknown for their strong predictive accuracy and inherent capability to address class imbalance through weighted loss functions and regularization. One of the key barriers in fraud detection is the highly imbalanced nature of transaction datasets, which can hinder model performance. To mitigate this, the study integrates strategic feature engineering and interpretability tools to enhance the detection of anomalous transaction patterns. The system’s effectiveness is assessed by using performance metrics such as precision, recall, and F1-score. Outcomes indicate that gradient boosting models outperform traditional methods like logistic regression and random forests, particularly in minimizing false negatives crucial in fraud prevention. The proposed approach strikes a balance between computational efficiency and real-world applicability, making it a viable solution for scalable, high-dimensional fraud detection environments.