Distributed Network Intrusion Detection Model Based on Multi-Feature Fusion Generative Adversarial Network and Federated Learning
Chong Zhao, Feifei Liao, Tao Yang, Rong Zeng · 2025
With the development of science and technology, distributed networks are widely used in various fields. However, their openness and complexity bring severe security challenges, especially network intrusion attacks, which may lead to serious consequences such as data leakage. Traditional intrusion detection models have difficulty dealing with the scarcity of attack samples, which leads to a bias towards normal samples during training and weakens the ability to detect intrusion behaviors. In addition, they have difficulty extracting complex features of distributed networks, which affects model performance. Finally, data privacy issues of heterogeneous devices limit data sharing and hinder the training of unified anomaly detection models. To address the above problems, this paper proposes a distributed network intrusion detection model based on multi-feature fusion generative adversarial network and federated learning (FL-MGAN). First, this paper designs a multi-feature fusion generative adversarial network (MGAN), which extracts multiple features through heterogeneous neural networks to enhance the ability of the generative adversarial network to generate complex heterogeneous features. In addition, the sub-nodes and central nodes of the distributed network are modeled as the client and server of federated learning respectively. The client trains and collects data locally, and the server aggregates and distributes model parameters, realizing the deployment of intrusion detection models while protecting data privacy. This paper demonstrates the effectiveness of the FL-MGAN through simulation experiments on the latest CIC IoT dataset 2023 dataset.