A lightweight mini-batch federated learning approach for attack detection in IoT

The Internet of Things (IoT) has recently gained importance in many fields. The use of IoT in different fields leads to an increase in a wide variety of network attacks. Many researchers have used artificial intelligence (AI) based approaches like machine learning and deep learning techniques to detect such attacks. Traditionally these AI techniques train an intelligent model at a cloud data center using the IoT network data gathered by different IoT devices. The sharing of IoT data with the cloud data center may affect the privacy of the user's sensitive data. The federated learning techniques can be used to generate an effective attack detection AI model that preserve the privacy of IoT users, but these mechanisms have higher computational complexities and require large number of federation rounds. So, to detect such attacks without compromising the privacy of IoT users, we propose a lightweight mini-batch federated learning mechanism, which is computationally efficient and requires minimum number of federation rounds to detect malicious attacks in an IoT network. The performance of the proposed mechanism was tested on benchmark IoT network datasets and the results show that the proposed mechanism achieves an overall attack detection accuracy of 98.85% with a false alarm rate of 0.09% and requires minimal computational resources.