Evolutionary Generative Adversarial Networks for Detecting Novel Polymorphic Ransomware

Abstract The increasing sophistication of polymorphic ransomware has presented significant challenges to traditional detection systems, which often rely on static signatures and heuristic patterns. The novel approach outlined in this article leverages a Generative Adversarial Network (GAN) framework enhanced with evolutionary algorithms to dynamically detect previously unseen ransomware variants. Through adversarial training, the generator produces synthetic ransomware samples that closely mimic real-world behaviors, allowing the discriminator to refine its detection accuracy over time. The evolutionary aspect introduces variation and selection mechanisms, enabling the model to continuously adapt and evolve in response to new threats. Experimental results demonstrated superior detection rates, recall, and adaptability, particularly in detecting highly obfuscated ransomware strains, compared to traditional models. Although computationally intensive, the model’s scalability and ability to generalize across a diverse range of ransomware families make it an ideal candidate for high-risk environments where detection precision is paramount.