A Credit Card Fraud Detection Method Based on Hybrid Sampling and Random Forest Algorithm

Credit card fraud poses a significant threat, leading to substantial financial losses and eroding consumer confidence. To combat this issue, we developed a robust fraud detection system leveraging advanced machine learning algorithms, includ- ing Mahalanobis Distance, SMOTE-ENN hybrid sampling, and Random Forest. Our approach begins with data preprocessing, where we handle missing values and encode categorical features. We then apply Principal Component Analysis (PCA) for feature extraction, enhancing computational efficiency. To address the inherent data imbalance, we implement a hybrid SMOTE-ENN strategy, which balances the dataset by generating synthetic minority class samples while eliminating noise. We evaluate multiple models—Random Forest, K-Nearest Neighbors, and a hybrid approach—by training them on the preprocessed data and assessing their performance using key metrics such as accuracy, precision, recall, and ROC AUC. Among them, the Random Forest model delivers outstanding performance. To facilitate real- world application, we developed a user-friendly web application that allows users to upload datasets and monitor fraud detection outcomes in real time. With precise detection capabilities, this accessible tool enhances fraud prevention efforts, making it a valuable asset for payment processing systems and financial institutions.