Intrusion Detection System using CNNs and GANs

Abstract

This study investigates the effectiveness of deep learning models, namely Generative Adversarial Networks (GANs), Convolutional Neural Networks with three layers (CNN-3L), and Convolutional Neural Networks with four layers (CNN-4L), in the domain of multi-class categorization for intrusion detection. The CICFlowMeter-V3 dataset is utilized to thoroughly evaluate the performance of these models and gain insights into their capabilities. The primary approach involves training the models on the dataset and assessing their accuracy. The GAN achieves an overall accuracy of 93%, while CNN-3L demonstrates a commendable score of 99.71%. Remarkably, CNN-4L excels with a flawless accuracy of 100%. These results underscore the superior performance of CNN-3L and CNN-4L compared to GAN in the context of intrusion detection. Consequently, this study provides valuable insights into the potential of these models and suggests avenues for refining their architectures. The conclusions drawn from this research indicate that CNN-3L and CNN-4L hold promise for enhancing multi-class categorization in intrusion detection systems. It is recommended to further explore these models with diverse datasets to strengthen overall comprehension and practical applicability in this crucial field.