Evaluating the Performance of Various SVM Kernel Functions Based on Basic Features Extracted from KDDCUP'99 Dataset by Random Forest Method for Detecting DDoS Attacks

The main goal of Denial of Service (DoS) attack is to restrict authorized users from gaining access to available services and resources or to prevent from processing the benign events. When a DoS attack is launched by multiple connected devices distributed over a network, the attack becomes a Distributed DoS attack (DDoS). The research community addressed various types of DDoS attacks in literature. DDoS attacks are very hazardous and difficult to resolve in real time. Each of these types of attacks has some key features that are identified to improve network security in real time. In this paper, an approach using Random Forest method is presented to extract the basic features from KDDCUP'99 dataset. With these features as the input, the proposed approach is smoothly extended for detection of new and unseen DDoS attacks with the assistance of nine support vector machine kernel functions namely Hyperbolic tangent kernel, Linear kernel, ANOVA RBF kernel, Spline kernel, Radial Basis kernel, Polynomial kernel, Laplacian kernel, Bessel kernel, and String kernel. The experimental study clearly shows that Laplace Dot support vector machine kernel (Laplacian kernel) function gives the paramount performance in terms of detecting seen and unseen DDoS attacks.