Developing robust machine learning models to defend against adversarial attacks in the field of cybersecurity
Taif Ayad Khaleel · 2024
Due to their vulnerability to malicious attacks, machine learning models utilized in cybersecurity applications necessitate robust safeguards. Despite previous research, there is still a lack of effective and practical protections for real-world circumstances. In order to tackle this issue, our work extensively investigates methods to enhance the robustness of machine learning models against malicious assaults. Our cutting-edge cybersecurity defensive tactics are derived on the extensive Edge-IIoTset cybersecurity dataset, specifically designed for Internet of Things (IoT) and Industrial Internet of Things (IIoT) applications. Our methodology integrates sophisticated techniques like adversarial training, input preprocessing, and evaluating the robustness of models. The proposed defensive measures have shown to be highly successful in mitigating the impact of hostile attacks, as evidenced by substantial empirical research. Specifically, as compared to the baseline models, our defensive model exhibits a notable 15% enhancement in accuracy. Our research demonstrates that safeguarding machine learning systems in real-world cybersecurity scenarios necessitates taking proactive actions. This will enable further advancements in the development of unique defense measures. Training methods could be enhanced by including adversarial assaults employing generative adversarial networks (GANs), random forest ensembles, and a variety of scenario-specific hybrid approaches. Assess their effectiveness in dealing with the issue of models being vulnerable to complex manual attacks.