Enhanced Credit Card Fraud Detection: A Machine Learning Approach

With the swift advancement in technology, credit cards are increasingly prevalent as substitutes for cash in daily transactions. This presents several opportunities for unscrupulous individuals to use credit cards fraudulently. 425,977 reports of credit card fraud were made in 2023, according to a Federal Trade Commission report. The credit card issuer ought to offer a service that shields customers from potential risks to guarantee their safety when using these cards. To address the problem of class imbalance, we resampled the dataset using the Synthetic Minority Oversampling Technique (SMOTE). The following variety of machine learning methods were applied to assess this framework using a soft voting mechanism: Combination 1 consists of KNN, Bagging, AdaBoost, and Logistic Regression, whereas Combination 2 consists of Random Forest, KNN, Bagging, and XGBoost. Furthermore, the usefulness of Logistic Regression, KNN, and AdaBoosting individually was evaluated to determine their stand-alone performance. This comparison served as a benchmark against which the ensemble models were evaluated. The models’ accuracy, F1 score, and Area Under the Curve (AUC) were used to evaluate them. Additionally, the outcomes of the machine learning models combined produced better results, with a total accuracy of 99.81% for combination 1 and 99.86% for combination 2 compared to the individual predictions of the models. However, Combination 2 outperformed Combination 1 in terms of F1 score, suggesting enhanced overall recall and precision performance.