Publication

Abstract : Credit card fraud is thereby a big challenge which results to enormous losses to both the individuals and the financial institutions. It is often the case that conventional measures of approach to fraud detection often fail to offer what is needed because they rely on broad characteristics of behaviour and not characteristics of the specific customer. The present work proposes a sophisticated fraud detection system equipped with a transaction-processing engine defining the fraud detection technology coupled with machine learning algorithms used to assess the credibility of transactions. To enhance the detection accuracy, the system incorporates Support Vector Classifier (SVC) and the MCC-based merchant profiles. To conduct a more realistic experiment, the synthetic dataset of 50,000 credit and debit card transactions with a minimal portion of fraudulent transactions was used. To assess the effectiveness of the Support Vector Machine (SVC), Decision Tree, and Random Forest algorithms, the study was carried out. The findings indicated that the use of MCC based profiles the category of Support Vector Machines (SVC) achieved an accuracy of 91 % which is remarkably higher than basic methods. To avoid the high risk of fraud and significant losses, the proposed system is designed to continuously check, verify, and process transaction requests. Further advancements will focus on enhancing the algorithm to adapt to the emerging techniques in fraudulent practices and enhance the security of transactions to a new level.