Network Intrusion Detection Based on Federated Learning with Inherited Private Models
Yanxiang Feng, Enyan Zhang, Yang Zhang, Zhen Zeng, Aili Wang · 2023
To solve the problem of insufficient and imbalanced data in Network Intrusion Detection (NID) in practical scenarios, which leads to low detection accuracy of the model. We propose a network intrusion algorithm based on the Federated Learning with Inheritance Private Model (FedPHP) , which applies Federation Learning (FL) to the field of NID to obtain a well-performing NID model in a privacy-preserving manner in scenarios with unbalanced small samples and multiple participants. This method uses oversampling to preprocess the data, and adds Inheritance Private Model (HPM), knowledge transfer and the matrix Adaptive motion estimation (Adam) algorithm in the training process. The experimental results show that each participant can still get a personalized model with high accuracy and strong robustness under the condition of providing limited data and not infringing the privacy of other parties. The average classification accuracy is 0.9874, and the performance is better than the traditional Deep Learning (DL) and FL network models.