Publication

Abstract : The rapid evolution of malware necessitates advanced detection methodologies to address the limitations of traditional signature-based approaches, which often fail against zero-day malware and polymorphic attacks. This study introduces a dynamic packet analysis framework, leveraging fine-grained network features such as packet length and time-to-live (TTL) for real-time detection. A hybrid architecture, combining Gradient Boosting and Random Forest models, improves detection capabilities to 95.7% with robustness against adversarial attacks. The model’s effectiveness in analyzing control flow traces outperforms traditional methods by 26.3% in detecting obfuscated malware and maintains an 83% success rate against adversarial threats. This approach not only sets a new benchmark in adapting to evolving cybersecurity challenges but also provides scalable and resilient malware detection