Granular computing framework for credit card fraud detection

Online credit card fraud detection presents significant challenges due to the dynamic and sophisticated nature of fraudulent activities. Fraudulent transactions are rare compared to legitimate ones, leading to highly imbalanced datasets that hinder traditional machine learning models from effectively identifying anomalies. Additionally, fraud patterns evolve rapidly as cybercriminals adopt new techniques, requiring detection systems to be adaptive and robust. The presence of irrelevant or noisy attributes in transactional data further complicates the process, potentially masking fraudulent activities and degrading model performance. Furthermore, striking a balance between minimizing false positives and detecting true frauds is a critical and ongoing challenge in this domain. This research proposes the use of a granular computing framework (GrCF) to enhance the performance of credit card fraud detection. This research highlights the strengths of the model by implementing a strategy based on three essential pillars. The model uses case-based reasoning (CBR) in conjunction with a mixed sampling technique to address the missing variables. The initial stage focuses on the distribution of classes within the dataset. The second step, the fuzzy rough set, refines the feature selection (FS) process by concentrating on the most important properties. In the third step, we develop the machine learning model by using the Boosted GWO (BGWO) method to maximize the hyperparameters of the Support Vector Data Description (SVDD). We have demonstrated through several tests that the proposed framework surpasses some current algorithms in accuracy and efficiency.