科研成果详情

The task of detecting credit card fraud represents a substantial hurdle in the realm of financial risk management, primarily due to the inherent imbalance of data between fraudulent (positive) and regular (negative) transactions. This imbalance can result in erroneous judgments and considerable financial losses. To mitigate this issue, we propose an effective classification strategy specifically designed for credit card fraud detection in this paper. The goal of our research is to regulate the rate of false negatives (Type I errors) while simultaneously minimizing the rate of false positives (Type II errors). We leverage the Neyman-Pearson paradigm and integrate several high-precision oversampling methods to equilibrate the training dataset, thereby enhancing the accuracy of fraud detection. To assess the efficacy of our proposed methodology, we perform exhaustive numerical analyses on a notably imbalanced European Credit Card dataset. Moreover, we modify the imbalance ratio to offer practical perspectives on the performance of our approach. The numerical outcomes demonstrate that our proposed methodology exhibits commendable performance.