Comparative Study on Deep Learning Approachesfor Fraudulent Transaction Detection

Abstract

Credit card theft has become a major problem in the financial world because of the huge rise in online shopping and the
complexity of fraud cases. A convolutional neural network (CNN) 1D could be useful in spotting fraudulent transactions, according
to recent research. As part of the preparation, the data will be standardised and validated. The Synthetic Minority Over-Sampling
Technique (SMOTE) will be used to balance the classes. Next, datasets are created for training, validation, and testing purposes.
Impressive results in domains such as recall, accuracy, and precision (including a 99.7 F1-score) are achieved by the proposed CNN
1D model, surpassing more traditional ML models like Support Vector Machine, Naive Bayes, and K-Nearest Neighbour.
Furthermore, the model's robustness and generalisability were demonstrated using learning rate optimization, ROC-AUC, and
confusion matrices. The results indicate that CNN 1D model is a highly predictable and scalable credit card fraud detection model
with significant better performance compared to the traditional methods in accuracy, sensitivity and scalability.