Bio-Inspired, Host-based Firewall
Lanier A. Watkins, James Ballard, Kevin Hamilton, Jay Chow, Aviel D. Rubin, William H. Robinson, Cleon E. Davis · 2020
In this research, we explore the viability of using a biologically-inspired design to develop a host-based next generation firewall. Many of the firewall designs in use today are still signature-based (with very few exceptions just released in 2020), which tend to struggle to protect against zero-day attacks and against encrypted traffic. By applying cellular membrane concepts such as endocytosis and ligand-receptor interactions along with machine learning, we created a feasible next generation firewall prototype capable of defending against zero-day threats even within encrypted traffic. Specifically, we built both supervised and unsupervised machine learning models, optimized their results, then we exported the models, and tested the exported models on zero-day malicious network traffic. Finally, we modified the open source Linux IP Tables firewall software to use our exported model as the packet filtering mechanism. The end result was a working host-based, supervised machine learning driven, next generation firewall prototype that demonstrated very promising results, nearly 100% of normal connections were passed and 77% of zero-day malicious connections were blocked.