Comparative Analysis of Machine Learning Models for Credit Card Fraud Detection Using SMOTE for Class Imbalance

Credit card fraud poses significant financial and security challenges, with negative consequences for consumers and financial institutions. An efficient, accurate detection system is essential. This study aims to determine which machine learning (ML)method performs best for detecting fraudulent credit card transactions by evaluating models such as Naive Bayes, Logistic Regression, k-NN, Decision Trees, as well as Random Forests, XGBoost, and AdaBoost. The models were evaluated using an open-access dataset from Kaggle, which includes actual payment activities conducted with credit cards by European cardholders in 2013. Due to data imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to enhance performance. Results indicate that Random Forest and XGBoost outperformed other models in terms of accuracy, F1 score, and the areas under the ROC (AUC) and precision-recall (AUPRC) curves. Specifically, Random Forest achieved an accuracy of 0.999, F1 score of 0.872, AUC of 0.978, and AUPRC of 0.871, while XGBoost reached an accuracy of 0.999, F1 score of 0.837, AUC of 0.983, and AUPRC of 0.867. In conclusion, Random Forest and XGBoost demonstrated superior performance, offering promising tools for effective credit card fraud detection. However, the use of 2013 data may limit the generalizability of results to more recent fraud patterns.