An optimal federated learning-based intrusion detection for IoT environment

Federated Learning (FL) allows the learning models in distributed systems to be trained by sharing the network data and model parameters. The attack patterns of attackers are frequently upgraded as well as the technology improves. Machine learning-based intrusion detection is familiar for cybersecurity in IoT networks. However, these traditional procedures mainly focus on training the machine learning model through specific data and parameters. This might reduce the detection performance of IDS as the system doesn't have insightful knowledge about the new attack patterns. Analyzing and detecting intrusions by analyzing diverse attack patterns is complex for machine learning algorithms. To overcome this, a federated learning-based intrusion detection approach is presented in this research work that trains deep learning classifiers in IoT networks through federated learning and detects different attacks. The Chimp optimization algorithm is used in the proposed work to select optimal features. Experimentations using the benchmark MQTT dataset validate that the FL-based IDS proposed in this research provides a maximum detection accuracy of 95.59% in detecting intrusions over traditional machine learning algorithms.