Federated Learning-Based Malware Detection for IoT Platforms

The proliferation of Internet of Things (IoT) devices, numbering in the billions, has unfolded in recent years without adequate security measures in place. This study presents a comprehensive exploration of the potential of federated learning to address IoT malware concerns while delving into the security intricacies inherent in this novel learning paradigm. We introduce a novel framework that leverages federated learning to detect malware threatening IoT devices. The framework’s effectiveness is evaluated using the N-BaIoT dataset, a meticulously designed representation of real-world IoT devices’ network traffic patterns influenced by malware instances. We provide both supervised and unsupervised federated models, exemplified by state-of-the-art perceptrons and autoencoders, with the ability to identify malware’s impact on both familiar and previously unseen IoT devices from the N-BaIoT collection. This empirical exercise highlights the robustness and generalizability of our federated models. Moreover, a pivotal facet of our study pertains to a head-to-head comparison between federated models and conventional methods. While traditional approaches confine participants to training models in isolation using their individual datasets, federated learning capitalizes on a diverse and extensive dataset, markedly amplifying model performance. It emerges from our findings that the federated models, all the while upholding data privacy tenets, yield results akin to their centralized counterparts.