Credit Card Fraud Detection in Imbalanced Datasets: A Comparative Analysis of Machine Learning Techniques

The growth of digital transactions has led to an increase in credit card fraud, creating financial risks for individuals and institutions. Fraud detection is challenging due to the class imbalance in transaction datasets, where fraudulent transactions often make up less than 1 % of the data. This study presents a comparative analysis of machine learning models applied to credit card fraud detection, focusing on addressing class imbalance using resampling techniques such as SMOTE and SMOTE-ENN [1], [3]. Traditional models, including Logistic Regression and Support Vector Machines, were evaluated alongside ensemble methods like Random Forest and Gradient Boosting Machines [2], [4], as well as deep learning models such as Long Short-Term Memory (LSTM) networks [5]. The results show that Random Forest with SMOTE-ENN achieved an AUC-ROC of 0.85, balancing precision and recall, while LSTM models paired with SMOTE-ENN delivered an AUC-ROC of 0.90. These findings demonstrate that ensemble methods and deep learning approaches, when properly optimized, can provide effective solutions for fraud detection. This study offers insights into the trade-offs between model accuracy, computational efficiency, and real-time applicability in fraud detection systems.