Investigating coevolutionary algorithms for finding Nash equilibria in cybersecurity problems

Abstract

Distributed Denial of Service (DDoS) cyberattacks continue to increase and cause disruptions in both industry and politics. As more critical information and services are provided through networks, it becomes more important to keep these networks available. However, since cyber-adversaries continuously change and adapt, stationary defense strategies do not effectively secure networks against attacks. We modeled attacker-defender interactions using competitive coevolutionary algorithms and investigated Nash equilibria within these cybersecurity problems. In particular, we examined and presented variations on two existing algorithms that look for Nash equilibria: NashSolve and HybridCoev. To compare these algorithms' performances against other existing heuristics, we considered multiple evaluation methods: the first calculates average fitness scores, the second creates a compendium of MEU, MinMax, and inverse Pareto front ratio scores, and the third utilizes Nash averaging. Although NashSolve and HybridCoev do not perform significantly better on average for either attacker or defender populations relative to other heuristics in these evaluations, they are able to produce strong individual strategies.