Detecting Distributed Denial of Service attacks using Recurrent Neural Network

Abstract

As the internet grows and diversity, attackers use various attacks to crash the servers and to stop specific sites. Multiple computers and multiple Internet connections are targeted by using distributed denial of service (DDoS) attacks. The aim of this paper is to identify the best algorithm among the selected algorithms (i.e., gradient descent with momentum algorithm, scaled conjugate gradient, and variable learning rate gradient descent algorithm.  In this study, the recurrent neural network was trained to check the accuracy and detection of DDoS attacks. The intention of this training was to allow the system to learn and classify the input traffic into the category. The proposed system's training was composed of three separate algorithms utilizing recurrent neural networks. The MATLAB 2018a simulator was used for training purpose. Moreover, clean the Knowledge Discovery Dataset (KDD) during design and include the values of protocols, attacks, and flags. The neural network model was subsequently developed, and the KDD was trained using Artificial Neural Network (ANN). The results of DDoS attacks’ detection were analyzed using MATLAB's ANN toolbox. The success rate of the variable learning rate gradient descent algorithm was 99.9% accuracy and the short timing was 2 minutes and 29 seconds. The variable learning rate gradient descent algorithm gives better results than gradient descent with momentum and scaled conjugate gradient algorithms.  In the state of the art, different algorithms have been trained in different neural networks and different KDD datasets by using selective DDoS attacks but in this research recurrent neural network was used for three different algorithms. In this research, we have used total of 22 attacks for detection of DDoS attacks’ accuracy.