Federated learning-based intrusion detection system for Internet of Things

Intrusion detection in the Internet of Things is becoming increasingly important as the number of connected devices grows. Machine learning algorithms can be applied to detect anomalies in large data sets, making them useful for identifying potential intrusions. However, traditional centralized learning techniques entail collecting data from end devices in one central device for training. Allowing a single entity to have access to vast amounts of personal data raises many security concerns as any issue experienced with the system can lead to widespread data leakage. To prevent these issues, it is critical to seek more secure alternatives such as federated learning. It enables multiple parties to collaborate on the same model without having to share the data between them. This process not only helps protect data privacy, but also reduces the risk of data leakage and improves training efficiency. In this paper, we propose a federated-based intrusion detection system. To better investigate the performance of the proposed model, we considered client-side evaluation whereby in the same round, the clients transfer the local models to the server which aggregates them in an updated global model. Then, the server transfers the updated global model to the clients for evaluation. The clients evaluate the global model locally and send back the results to the server to be aggregated using metric aggregation function. The experimental results show that the proposed federated-IDS achieves a high detection rate.