PERFORMANCE ANALYSIS OF DEEP LEARNING APPROACH FOR CLASSIFYING DDOS ATTACK FROM BENIGN NETWORK TRAFFIC

Abstract

Cyber security has become a great issue in this technological world. There are several types of cyber-attacks that are present, where Distributed Denial-of-Service (DDoS) is one ofthe most common attack type in the cyber world. Researchers are doing their best to find a solution to get rid of DDoS attacks. With the advancement of technology day by day, millions of people across the world are relying on the internet. People are using internet in every field of life from the very basic home task to the academics ofusers are increasing day by day, security issues are also increasing. DDoS has grown significantly than normal. DDoS attacks frequency is doubled in every year but due to COVID-19 pandemic, as everything is shifted on internet. To identify and to take measures against DDoS attacks has become a necessary task. There is a need to make a system intelligent enough to detect the difference between the legitimate request and DDoS attack request. Blocking the traffic is not a solution. It is important to develop a technique which is intelligent enough to distinguish the normal and malicious traffic. research. As the number more There are many solutions available up till now. Researchers are using different techniques to get rid ofthis problem. In this research, three different approaches are used to check which one is better for cyber security dataset. The dataset used is CICDDoS 2019 comprises of different DDoS attack types. The first approach is Machine Learning approach in which Random Forest algorithm are used. Second approach consists of ANN (Artificial Neural Network) and CNN (Convolutional Neural Network). The performance of CNN and RF is almost same. Accuracy obtained by using of all the three approaches are better. In some of the attack classification, the accuracy is increased up to 99.9%. Whereas ANN algorithm has an average performance for cybersecurity dataset. There are many anomalies occurred in the performance ofANN. The performance parameters include Accuracy, Training Time, Testing Time and Confusion Matrix. CNN takes more time in training than RF but there is a very less chance of any