Machine Learning Strategies for Securing Financial Transactions against Risks

The increase in online credit card fraud requires robust detection techniques. Innumerable studies suggest innovative fraud detection methods viz. anomaly detection, deep neural networks, and machine learning algorithms. Methods viz. IoT2Vec, swarm intelligence, random forest, and graph models improve fraud detection accuracy in earlier research. This project focuses on transaction classification using various ML algorithms to improve accuracy and efficiency. Evaluation shows XGBoost outperforms with perfect ROC scores of 1.0 on train data and 0.98 on test data. Yet, Logistic Regression with SMOTE excels due to simplicity and interpretability, achieving ROC scores of 0.99 on train set and 0.97 on test set, making it the optimal choice despite XGBoost's promising results. To fight online credit card fraud, a varied approach from anomaly detection to ML is crucial. Though XGBoost performs well, Logistic Regression with SMOTE is more efficient.