A Novel Deep Learning Framework for Credit Card Fraud Detection

Credit card transactions are the norm in most online sales and purchases, which are growing every day. While this is great for the customer, it also raise the risk of fraudulent charges made through online credit cards. Credit card fraud detection has grown exponentially in response to this security concern in recent years, with certain nations seeing an astronomical spike in fraud as a result. Despite the fact that ML-based solutions have been suggested in recent research as a means of recognising fraudulent credit card transactions, the imbalance of classes in any particular dataset means that their discovery scores still have room for perfection. Very few methods have shown remarkable performance across several datasets. To address the text data issue, this study employed a deep learning (DL)-based strategy that was trained on the Kaggle dataset. We offer a new text2IMG conversion method that can produce tiny images. To fix the problem of class imbalance, the pictures are fed into a convolutional neural network (CNN) that uses the inverse frequency approach to determine class weights. The suggested system's validity and robustness were confirmed using DL and ML methods. By exploiting the deep features of the proposed CNN was able to attain an accuracy of 99.87%.