Detection of DDoS Attack Based on Deep Neural Network with various Number of Features

Abstract

Distributed Denial of Service (DDoS) attacks have become an effective threat to the reliability and availability of the services of the internet in last decades. The effectiveness of utilizing Deep Neural Networks (DNNs) for DDoS attack detection is investigated in this paper. We implemented a reliable detection system that analyzes network traffic data to spot any DDoS activities. A multi-layer perceptron model trained on a dataset containing five different forms of DDoS attacks and normal traffic is used in this method. Also, three cases of varous number of features was investigated to extract the optimal number of features that can be used for detection of DDoS attacks. To improve accuracy, a great deal of testing was done on the model's architecture using various hyperparameters and training procedures. With a 96.5% detection rate, the DNN results showed a high degree of accuracy. This demonstration highlights the ability of deep neural networks to identify DDoS attacks in the midst of regular traffic. The six-category classification enhances detection granularity and facilitates the application of more specialized and successful mitigation techniques. Given the great precision attained, DNNs have the potential to be an essential part of real-time detection systems, providing a major advancement over conventional techniques.