Online Payment Fraud Detection using Machine Learning

Online payment fraud has become a major concern for financial institutions, driving the need for effective and efficient detection mechanisms. In response, this work focuses on an innovative method for detecting online payment fraud by leveraging the capabilities of Extreme Gradient Boosting (XGBoost) along with Bayesian optimization for fine-tuning hyperparameters. This approach tackles the challenge of severe class imbalance. XGBoost serves as the primary model for classification, with its performance enhanced through Bayesian hyperparameter optimization. To address the class imbalance, strategies such as instance weighting and threshold adjustment are employed. Our evaluation focuses on metrics such as F1-score, accuracy, and precision. This holistic approach holds great promise for enhancing fraud detection systems in financial institutions, striking a balance between detection accuracy and enabling continuous improvement.