Efficient Feature Selection On Adversarial Botnet Detection
Raqeebir Rab, Farsha Bindu, Sheikh Sanjida, Nafiza Tabassoum, Tamima Binte Wahab, Anonnya Ghosh · Annals of Computer Science and Information Systems · 2023
Botnet attacks now pose a significant hazard to the security and integrity of computer networks and information systems.However, due to technological advancements and the proliferation of malware, machine learning-based Intrusion Detection Systems (IDS) are incapable of protecting against such cyberattacks.IDS cannot detect novel bots because the vast majority of them are programmed systems.Keeping IDS upto-date with new malware varieties is, therefore, a crucial task.In this paper, we employ Generative Adversarial Networks (GANs) in which two neural networks compete and endeavor to outperform each other, which will serve as self-training for IDS.Our paper's primary objective is to develop an IDS capable of detecting novel malware with fewer attributes in real-time.To accomplish this, we present a method for feature selection that trains GAN models with a minimal subset of features so that the Generator can generate similar false bots with fewer features and the discriminator's ability to identify fake data improves.We used Pearson Correlation, the Wrapper method, and Mutual Information to select the best training model characteristics.The experimental evaluation suggests the GAN model in conjunction with Mutual Information is superior at detecting novel malware.