Federated Learning Based Approach to Intrusion Detection
Evgenia Sergeevna Novikova, Sergey A. Golubev · 2023
The use of federated learning (FL) for building intrusion detection systems solves the problem of using data with limited access, which cannot be made publicly available or shared with the third parties, to form analytical models for intrusion detection. This approach expands the variety of data used to train intelligent models, thereby increasing the detection rate of the heterogeneous attacks. In this paper, an intrusion detection system (IDS) architecture based on federated learning principles is considered. Taking into account that application of FL establishes certain requirements to the computational resources of the IDS components and network bandwidth capacity, a methodology for evaluating the FL based IDS performance considering attack detection rate in context of the training computational performance is proposed.