Research on network intrusion detection based on differential privacy federated learning

Federated Learning (FL) is a distributed machine learning technique that effectively protects client data from exposure to competitors. However, FL is still at risk of privacy breaches through the parameters uploaded by clients. To address this issue, researchers have developed Differential Privacy Federated Learning (DP-FL) strategies. These strategies introduce artificial noise to obscure parameter information, thereby safeguarding privacy. Despite this, DP-FL often comes at the cost of reduced model performance. To balance model accuracy and privacy security, we propose a Dynamic Differential Privacy Federated Learning (NIDS-DDPFL) method for network intrusion detection. This approach utilizes a dynamic privacy budget to add artificial noise, perturbing client parameters in a way that enhances privacy protection. We conducted extensive experiments to evaluate a range of parameters. The results demonstrate that our method significantly outperforms traditional DP-FL algorithms on the NSL-KDD dataset. Specifically, our NIDS-DDPFL method improves model performance while ensuring robust privacy security. In summary, our dynamic approach to differential privacy in federated learning provides a promising solution to the trade-off between model accuracy and privacy, making it particularly suitable for applications in network intrusion detection.