Cyberattack Risk Classification for Detecting Intrusions through N euro- Fuzzy Inference based Generative Adversarial Networks
Martha Torres, Sai Mounika Errapotu, Virgilio González · 2024
Intrusion detection systems playa significant role in securing the communication networks from adversarial attacks. Recent trends have seen an increase in the exploitation of network vulnerabilities to gain access to network. To address these issues Machine Learning/Artificial Intelligence based frameworks are being integrated into network defense techniques to build intrusion prevention/detection systems capable of scaling with network size while ensuring strong security against cyber-attacks. However, such intrusion detection systems' training procedure is becoming challenging due to the diversity in cyber and cyber physical attack vectors, as well as due to the lack of data, patterns, and parameters. In many applications, the primary challenge in developing an effective AI based intrusion detection system is the lack of adequate cyber-attack data needed for training, which directly affects the accuracy of alerts generated since the training attack dataset is not a full representative of network dynamics. To address these challenges, we focus on Generative Adversarial Networks (GANs) based Intrusion Detection System (IDS) capable of creating realistic attack samples needed for training as well as augmenting missing data. This work specifically focuses on the classification problem in the discriminator side of the GANs based IDS to improve accuracy in training. This paper presents an innovative neuro fuzzy inference injection technique based on the Mamdani approach. The proposed three phase approach that includes data preprocessing, fuzzy inference-based training, and rule evaluation procedure for GAN s based IDS is validated and tested for improving model performance, reliability, and efficiency on IEEE loT dataset.