Abstract

Social networking platforms have become prime targets for bot-driven cyber threats. The unpredictable expansion of social networking platforms has resulted in bot-driven cyber threats, which includes spreading misinformation, spam, phishing, and automated account takeovers. Traditional machine learning models are not very effective as they are susceptible to combative attacks. This paper proposes a novel robust detection framework using adversarial machine learning (AML) techniques to enhance the robustness of bot detection systems. Attacks involving evasion and Poisoning, directed at ML-based security systems, have been observed, and some defensive strategies help counter these types of attacks. Experimental results on benchmark datasets demonstrate significant improvements in detection accuracy compared to traditional models.