Botnet Intrusion Detection Method based on Federated Reinforcement Learning

Aiming at the isolated island problem of botnet datasets, botnet intrusion detection method based on federated learning is proposed, which needs not train data centrally, but only needs to aggregates the model parameters trained by clients using local data through the central server. However, traditional federated learning algorithms tend to be characterized by non-independent and homogeneous distributions, so we proposed a botnet intrusion detection method based on federated reinforcement learning in this paper, which uses deep reinforcement learning algorithm to intelligently select nodes to participate in training on the basis of federated training to offset the impact of data heterogeneity, thereby improving the training speed and reducing communication costs. In the experimental part, the underlying model Gate Recurrent Unit (GRU) and Long Short Term Memory (LSTM) are experimentally compared and evaluated.