USING GENERIC ALGORITHM TO SYSTEMATICALLY CHOOSE PARAMETERS IN DETECTING DISTRIBUTED DENIAL OF SERVICE (DDoS) ATTACK

Distributed Denial of Service (DDoS) attacks are deliberately formed by an attacker (master) to overwhelm targeted server by recruiting numerous online computers (bots) to send large amount of useless requests to prevent it from responding to other legitimate users. Research works have confirmed that DDoS attacks take place only where multiple bots are instructed to overwhelm single server with series of useless packets. Numerous machine learning-based methods have been proposed and developed to detect and prevent these attacks; in most cases, these approaches are successful in detecting many DDoS attacks. However, some of these approaches are slow while validating normal network packet therefore increasing the computational time of the process. Sufficient literatures have shown that, the problem highlighted occurred due to non-normalized dataset prior to employ detection mechanism and selecting features of network packets directly from the dataset increases computational time and unnecessary delay normal network packets to reach their destinations in time. Therefore, this research work has classified the network packets into four groups with respect to their sources and destinations, these are: (i) Single Source to Single Destination (ii) Single Source to Multiple Destinations (iii) Multiple Sources to Multiple Destinations (iv) Multiple Sources to Single Destination. In order to tackle highlighted problems an optimal solution is needed, by obtaining network packet features that are capable to perform best in detecting attack packets using an existing mechanism. Group (iv) packets are selected, and relevant features are carefully selected from group with the help of decision tree in WEKA application. Selected features are used to distinguish between normal and attack packets with the help of Hybrid Modified Grasshopper Optimization Algorithm with Generic Algorithm (HMGOGA) model. Finally, the research validates the effectiveness of the method and compared it with some related works. The results show that the proposed method had minimized computational time by 81.66% and improved detection accuracy by 99.82%.