Comparative Evaluation of GAN, CGAN, and ACGAN for Intrusion Detection in Cybersecurity
Chalerm Klinkhamhom, Pongpisit Wuttidittachotti, Pongsarun Boonyopakorn · 2025
This study investigates the effectiveness of Generative Adversarial Networks (GANs), including Conditional GAN (CGAN) and Auxiliary Classifier GAN (ACGAN), for synthesizing attack traffic in network intrusion detection systems (NIDS). The proposed framework evaluates the representational quality of synthetic data generated from each model by training classifiers exclusively on fake attack samples and testing them on real-world data. Two publicly available benchmark datasets, NSLKDD and UNSW-NB15, are used to ensure realistic and diverse evaluation conditions. Experimental results show that CGAN and ACGAN outperform the vanilla GAN across key metrics, including Accuracy, F1-score, and AUC. CGAN achieves the highest AUC of 0.6592 on UNSW-NB15, while ACGAN yields the best F1-score in detecting minority classes on NSL-KDD. These findings highlight the potential of conditional GAN-based models to generate high-quality, class-aware attack data, offering a promising approach for addressing data imbalance and improving machine learning-based intrusion detection systems.