Generative Adversarial Networks for Dynamic Cybersecurity Threat Detection and Mitigation

Abstract

The increasing complexity and dynamism of cyberattacks, such as ransomware, phishing, and denial of service, demand advanced solutions that overcome the limitations of traditional methods, such as support vector machines and decision trees. This study proposes a generative adversarial network (GAN)-based model to enhance the detection and mitigation of dynamic cybersecurity threats by improving adaptability and robustness in real-time scenarios. The model is designed to detect anomalies in network traffic and generate malicious synthetic patterns to strengthen system defenses. The model was trained and tested using publicly available datasets, CICIDS2017 and UNSW-NB15, and an experimental environment simulating corporate networks with 50 interconnected devices generating realistic traffic to evaluate its effectiveness. The results demonstrate that the GAN-based model achieved an average precision of 92%, an F1 score of 91%, and robustness against noise of 89%, significantly outperforming traditional approaches. The key novelty of this work lies in integrating noise robustness and generalization as primary evaluation metrics, along with the ability to generate real-time countermeasures, making it a more resilient solution in dynamic cybersecurity environments. These findings suggest that the proposed approach offers a significant advancement in the field, enabling better adaptability to evolve threats and improving security frameworks in complex network infrastructures.