Search

In this paper, we propose a new DDoS attack detection mechanism based on federated learning that employs dynamic thresholds to cope with the fluctuation of variable rate DDoS attacks. The performance of this detection mechanism is analyzed in terms of traffic classification, performance of verification module, accuracy and loss value. Experimental results show that the method has an accuracy of 99.83% in detecting regular Benign traffic. In burst attack scenarios, the technique significantly improves detection accuracy for all 10 common DDoS attack types. In a sustained attack environment, the intrusion detection system trained based on the DDoS model has the most minor performance degradation, and the average detection accuracy for all types of DDoS attacks still exceeds 90%. Compared with the traditional SVM model, the DDoS attack detection model based on federated learning has a significant performance advantage with Loss and Acc parameters of 0.1 and 0.9, respectively.