PPFL-IDS: Privacy-Preserving Federated Learning Based IDS Against Poisoning Attacks

In view of the existing federated learning can only encrypt the model to ensure data privacy, but can not guarantee the correctness of the uploaded model, this paper proposes an anti-poisoning attack intrusion detection scheme based on privacy-preserving federated learning. First, an anti-poisoning attack algorithm based on the encryption model is designed, and a comprehensive anti-attack model is proposed. On this basis, the model defines the defense strategy, and objective function, and introduces the poisoning rate into the objective function to make the model take into account the availability and concealing of attack. The privacy of local data is protected while the intrusion detection model based on knowledge sharing among islands is constructed. The experimental results show that the proposed scheme can greatly improve the robustness of the detection model, and its accuracy rate can reach 83.11% even after being poisoned, and the detection performance has not significantly decreased compared with that before being poisoned.