A Comparison of Data Balancing Techniques for Credit Card Fraud Detection using Neural Network

This research study has employed an unequalled credit card dataset and evaluated several approaches of data balancing by using a basic neural network-based model with only one hidden layer of size 32. Even with a simplified design, the supervised machine-learning model has produced optimal results in contrast to sample-based approaches for the over-sampling of data balance. The proposed model based on random over-sampling performs better than other two model based on over-sampling approaches (SMOTE, SMOTETomek).