Detecting Cyber Threats and Enhancing Security in IOT Devices: A Federated Learning Approach

The aim of the study is to detect cyber threats in Internet of Things (IoT) devices using federated learning vis-a-vis centralized learning. Unlike centralized learning, which requires data transfer to a central server, federated learning offers a decentralized approach, processing data locally on devices, thus potentially minimizing privacy risks and addressing scalability concerns. The project encompasses a feed-forward neural network as the base model for both the centralized and federated settings for easy comparison. The findings, based on extensive trials, revealed that while centralized learning delivered robust performance, federated learning showcased even better performance with enhanced privacy features, reduced data transmission costs, and demonstrated exceptional scalability. Moreover, when examining state-of-the-art techniques, the developed federated learning model achieved a commendable accuracy of 98%, rivaling the performance of centralized models. Given the escalating concerns over data privacy and the growing scale of IoT networks, this research underscores the viability of federated learning as a more privacy-preserving and scalable solution for threat detection in IoT devices. It contributes to the applications of machine learning in IoT, primarily in strengthening the security of IoT networks with federated learning.