Optimizing Fraud Prevention in Financial Transactions using Scalable Machine Learning Models based on Credit Card Data

Abstract

Due to the increasing number of cyberattacks and frauds, especially in credit card transactions, fraud prevention in financial transactions has been even more important.  The inherent difficulties of preventing fraud in financial transactions necessitate complex machine-learning in order to detect the frauds effectively and efficiently.  This paper employs deep learning architectures, Fully Connected Neural Networks (FNN) and Convolutional Neural Networks (CNN), in order to classify fraudulent transactions based on European Customers Credit Card Transactions dataset. Various classification methods will be compared and contrasted in this study. With an accuracy of 99.87% and 99.61 over 30 trials, respectively, the suggested FNN and CNN considerably outperformed conventional models. Contrarily, during 10 experiments, the FNN achieved a high accuracy of 99.82 percent, and the CNN attained that of 99.81, indicating the stability/sturdiness of CNN. Although traditional models could be effective, their low recall and precision raised the chances of false negative results. Further evidence of the deep learning-based approach's dependability in real-world fraud detection situations was its enhanced precision, recall, F1-score, and AUC-ROC values. These results indicate an effectiveness of DL methods in the reduction of financial risks and increasing cybersecurity systems.