Deep Neural Networks Hyperparameter Optimization Using Particle Swarm Optimization for Detecting Frauds Transactions

Abstract

The recent explosions and developments of new technologies have changed our lives, and that was shown by the quantity of information shared, posted, and stocked in big companies like Facebook, Google, Amazon, and so forth. Millions of transactions are made by cardholders every year to buy online using credit card as a mobile wallet or for sample payment and that make credit card transaction more frequent today. The developments of communication technologies and E-commerce have made credit cards the most ordinary methods of payment for both online and regular purchases. As a result, millions of online transactions are subject to various types of fraud. So security in this system is required to prevent fraudulent transactions. In this direction, researchers to detect this fraud invent many approaches. Traditional techniques cannot detect sophisticated fraudulent. Furthermore, analysis of cardholder behaviors or static risk management rules of the frauds have never stopped the fraudsters to commit their crimes. However, artificial intelligence techniques such as deep learning and machine learning have been able to handle these issues. This paper proposes an approach to detect fraud transactions by optimizing Deep Neural Networks (DNNs) hyperparameters using Particle Swarm Optimization (PSO) as optimization methods and compare them with the grid search (GS) method. The results obtained in terms of precision, accuracy, recall, F1-score, Time, and Area under the Curve (AUC) have shown that the PSO can generate better solutions in a short time in comparison with the GS method.