SynKGen: A kernel PCA-Based oversampling method for enhanced credit card fraud detection

Credit card fraud detection is a growing challenge in the financial domain due to data imbalance, where fraudulent transactions are minimal compared to legitimate ones. This study presents SynKGen, a data augmentation method using Kernel PCA with Gaussian perturbations to generate synthetic samples of the minority class, contrasting it with ADASYN and SMOTE. By introducing variance analysis with controlled perturbations in the minority class, the proposed approach mitigates the risks of overfitting associated with traditional interpolation-based techniques. Four classifiers, XGBoost, RandomForest, AdaBoost and VotingClassifier, were evaluated using the original data set and variants with data augmentation. The RandomForest classifier achieved the best performance when using data generated with SynKGen (accuracy: 0.9949, precision:0.9899) outperforming the results obtained with ADASYN and SMOTE. Experimental results demonstrate that SynKGen improves the effectiveness of credit card bank fraud detection. These findings highlight the importance of data augmentation strategies to optimize classifier performance in financial contexts with unbalanced data.