Federated Learning Model for Network Intrusion Detection with Generative Adversarial Network and Explainable AI in Edge Environments

Abstract

Accelerated growth of the IoT in recent years has been driven by advancements in wireless transmission and processing. Various edge devices have appeared within IoT networks, such as smartphones, smart military equipment, smart cars, and intelligent applications. These applications are gaining popularity as they effectively tackle real-time issues. However, IoT networks are susceptible to a variety of threats because of their low processing power and resources. Therefore, in order to preserve data security and maximize network performance, efficient intrusion detection methods are required. This paper presents a federated learning model for detecting network intrusions using GAN and explainable AI within edge environments. The dataset is processed by addressing missing values and standardizing the data to enhance detection accuracy. Missing information is filled using a Recurrent Noise Injector GAN (RNI-GAN) technique, and the data is normalized with a robust scaler. Network Intrusion Detection over Generative Adversarial Network (N-GAN) is then used to identify suspicious/malicious network activity with respect to each edge device. The updates of each edge device are sent to a central server where they are aggregated to produce a better global detection model. The global model can explain its decisions through the SHAP method and make the process of detection transparent and comprehensible. The updated global model is transmitted to all the edge devices. This enables every device to improve its threat detection capabilities by relying on information containing the entire network. The system operates continuously, enabling the global model to get better over time by learning from new and changing data collected at each edge device. The proposed model attained the performance metrics of 97.19% accuracy, PPV of 97.14%, 2.81% error and F1_score 97.16% with other existing algorithm. Thus the proposed model is well to detect the intrusion detection in networks.