COMPARATIVE PERFORMANCE OF MACHINE LEARNING MODELS FOR CREDIT CARD FRAUD DETECTION ON IMBALANCED DATA: A STUDY USING SMOTE AND THE KAGGLE EUROPEAN DATASET

Abstract

Credit card fraud continues to be a major challenge in financial systems, causing substantial monetary losses and undermining trust in electronic transactions. Detecting fraudulent activities is complicated by the highly imbalanced nature of transaction datasets. This study evaluates the performance of four machine learning models including LinearSVC (Fast SVM), Logistic Regression, Decision Tree, and Random Forest using a Kaggle credit card dataset. Data preprocessing involved handling class imbalance using the Synthetic Minority Oversampling Technique (SMOTE) and standardizing feature values to improve model training stability. Each model was trained and assessed using metrics such as accuracy, precision, recall, F1-score, and AUC–ROC, with confusion matrices and ROC curves providing visual evaluation. Experimental results demonstrate that LinearSVC achieved the highest balance between precision (0.9983) and F1-score (0.9881), while Random Forest achieved near-perfect accuracy (0.9999) but slightly lower precision (0.8800) and recall (0.9002), highlighting trade-offs between overall accuracy and fraud detection sensitivity. The findings emphasize the importance of careful model selection and preprocessing in imbalanced financial datasets. Future work will investigate the integration of deep learning models and real-time fraud detection frameworks, as well as validation on additional datasets from diverse financial environments to enhance model robustness and generalizability.