A Novel Approach Integrating Autoencoders and ESMOTE-GAN for Credit Card Fraud Detection

An innovative Credit Card Fraud Detection Algorithm that merges two potent techniques, Autoencoders and Ensemble Synthesized Minority Oversampling Techniques using GANs (ESMOTE- GAN), to effectively address the pressing issue of credit card online payment fraud. The fraud detection, traditional models encounter significant challenges due to the skewed nature of transaction datasets, with the majority of entries representing legitimate transactions. Our approach tackles this challenge head-on by elevating the quality of feature representation and diminishing noise through the application of Autoencoders. By reducing the dimensionality and highlighting meaningful patterns within the data, Autoencoders enable our algorithm to capture essential information while ignoring extraneous variables. Furthermore, we strategically employ ESMOTE-GAN to rectify dataset imbalances. This entails generating synthetic minority-class samples, thus rectifying the skewed distribution without compromising on model accuracy and improvements of performance and the detection rate. Furthermore, our technique was rigorously evaluated by comparing it against a range of classifiers, such as Random Forest (RF), k-Nearest Neighbours (KNN), Support Vector Machines (SVM) and XG Boost. We streamlined the decision-making process by integrating the probability outputs from our trained classifiers through a weighted voting system. The results clearly indicate substantial enhancements in both overall performance and detection rates. Therefore, our model’s heightened detection capabilities not only enhance security but also lead to a significant reduction in the resources needed for manual analysis, presenting a more efficient and cost-effective solution.