Performance Evaluation of an Intelligent and Optimized Machine Learning Framework for Attack Detection

Abstract

In current decades, the size and complexity of network traffic data have risen significantly, which increases the likelihood of network penetration. One of today's largest advanced security concerns is the botnet. They are the mechanisms behind several online assaults, including Distribute Denial of Service (DDoS), spams, rebate fraudulence, phishing as well as malware attacks. Several methodologies have been created over time to address these issues. Existing intrusion detection techniques have trouble in processing data from speedy networks and are unable to identify recently launched assaults. Ineffective network traffic categorization has been slowed down by repetitive and pointless characteristics. By identifying the critical attributes and removing the unimportant ones using a feature selection approach could indeed reduce the feature space dimensionality and resolve the problem.Therefore, this articledevelops aninnovative network attack recognitionmodel combining an optimization strategy with machine learning framework namely, Grey Wolf with Artificial Bee Colony optimization-based Support Vector Machine (GWABC-SVM) model. The efficient selection of attributes is accomplished using a novel Grey wolf with artificial bee colony optimization approach and finally the Botnet DDoS attack detection is accomplished through Support Vector machine.This articleconducted an experimental assessment of the machine learning approachesfor UNBS-NB 15 and KDD99 databases for Botnet DDoS attack identification. The proposed optimized machine learning (ML) based network attack detection framework is evaluated in the last phase for its effectiveness in detecting the possible threats. The main advantage of employing SVM is that it offers a wide range of possibilities for intrusion detection program development for difficult complicated situations like cloud computing. In comparison to conventional ML-based models, the suggested technique has a better detection rate of 99.62% and is less time-consuming and robust.