Large-scale generative adversarial networks for simulating cybersecurity offensive-defensive confrontations

Recent advances in artificial intelligence have significantly strengthened cybersecurity, particularly in network intrusion detection and adversarial attack modeling. However, most existing security frameworks focus primarily on defensive mechanisms and lack the capability to simulate realistic cyberattacks for robust system evaluation. To address this limitation, this study proposes an offensive–defensive cybersecurity framework based on a Big Generative Adversarial Network (BIG-GAN) to model and analyze advanced network attack patterns. The framework utilizes the Canadian Institute of Cybersecurity Intrusion Detection System 2017 (CIC-IDS2017) dataset and incorporates preprocessing steps including data cleaning, feature encoding, and min–max normalization. The BIG-GAN architecture effectively learns traffic distributions and generates synthetic adversarial samples that closely mimic real malicious activities. Similarity analysis demonstrates high fidelity between generated and real traffic, particularly for Flow Bytes/s (0.998697) and Flow Packets/s (0.996923), while temporal features such as flow duration and packet rates show moderate similarity (approximately 0.92). For defensive evaluation, the model achieves 98.63% accuracy, 98.12% precision, 97.94% recall, and an F1-score of 98.03% in detecting malicious traffic. These findings confirm the framework’s effectiveness in enhancing cybersecurity resilience.