Abstract

Federated learning is a way to allow privacy-preserving intrusion detection systems (IDS) without sharing raw data. But the traditional methods, such as Federated Average(FedAvg), do not work well on Non-Independent and Identically Distributed(non-IID) and imbalanced datasets. This study presents an F1-Guided Softmax Aggregation method, which is achieved by assigning weights to the clients using the F1-scores to enhance the global model. The results obtained from the experiments conducted on the CICIDS2017 data set demonstrate that the performance of the proposed model is more accurate, precise, recall, and with better convergence stability. The proposed algorithm outperforms the other two, FedAvg and Scaffold algorithms, with almost 3 percent F1 score and reduces the number of false negatives effectively.