MULTI-CLASSIFIERS SYSTEM FOR CREDIT CARD FRAUD DETECTION

Abstract

The business of issuing credit cards is extremely important to the functioning of the economy since it facilitates the use of a straightforward method of payment in a variety of contexts, such as online banking, commercial transactions, and financial dealings. Nonetheless, using credit cards is also associated with fraudulent activity and non-payment, both of which constitute a considerable danger to customers and the business as a whole. Detecting and preventing fraudulent activity involving credit cards is complex and time-consuming because of the ever-changing nature of fraudulent and expected behavior and the unequal class label and overlapping of class instances inside the data sets. The unbalanced class distribution of the data sets is one of the most significant obstacles in the way of detecting fraudulent activity on credit cards. Typically, the number of fraudulent cases is substantially lower than those that do not involve fraud. This might result in biased models that have a high level of accuracy when applied to the majority class but perform badly when applied to the minority class. The overlapping of class samples is another obstacle in the way of detecting fraudulent activity on credit cards. This occurs when fraudulent and non-fraudulent transactions can appear to be very similar to one another, making it difficult to differentiate between the two. This study intends to resolve these issues by comparing the performance of single and multi-classifier methods. The study underlines the influence of oversampling and undersampling techniques on single-classifier methods and stresses the significance of selecting an appropriate classifier algorithm. The results indicate that multi-classifier methods, particularly COPOD + RFC and IForest + RFC, can substantially improve credit card fraud detection compared to single-classifier methods. These results demonstrate the potential advantages of integrating multiple unsupervised and supervised learning algorithms to improve credit card fraud detection while decreasing false positives. Overall, the study emphasizes the significance of employing a combination of machine learning techniques to resolve the difficulties of credit card fraud detection. The proposed method can assist credit card companies in accurately and efficiently iden- 3 tifying fraudulent activities, thereby reducing the risk of financial loss and enhancing customer confidence.