Credit card fraud poses a significant challenge to financial institutions, necessitating models that can detect rare and complex fraudulent patterns in highly imbalanced datasets. This study introduced a hybrid approach that integrated one-class support vector machine (OCSVM) for anomaly removal, random forest (RF) for classification and genetic algorithm (GA) for hyperparameter optimisation. The method was evaluated on the publicly available PaySim dataset, which simulates real-world financial transactions. Model performance was assessed using F1-score, accuracy, precision, recall, average precision-recall curve and area under the receiver operating characteristic curve. The proposed model achieved superior results compared to baseline methods, with an F1-score of 90.74% and accuracy of 99.95%, along with notable improvements in precision and recall. These findings demonstrate that combining anomaly detection with optimised classification significantly enhances fraud detection performance. The proposed framework can serve as a robust tool for improving fraud prevention in financial applications.
Alamri et al. (Fri,) studied this question.