Federated Defense: A Privacy-Preserving Deep Learning Model for IoT Malware Detection

As the Internet of Things (IoT) continues to expand, securing the vast network of IoT devices, particularly in Machine-to-Machine (M2M) communication, has become a critical concern. Traditional security approaches often fall short, particularly in protecting privacy and ensuring scalability across IoT systems' diverse and vast landscapes. This paper introduces the Federated Defense Model, a novel approach that harnesses federated learning (FL) to enhance IoT malware detection while preserving user privacy. Unlike centralized models, the proposed FL framework processes data locally on IoT devices, avoiding transmitting sensitive information and reducing bandwidth demands. We developed and evaluated a lightweight, one-dimensional convolutional neural network (CNN) optimized for the typical environment of IoT devices. Using the IoT-23 dataset, a collection of labeled network traffic representing various malware and benign scenarios, the experiments demonstrate that the proposed Federated Defense Model achieves superior accuracy and precision compared to the signature, heuristic, and traditional machine learning-based security models. Moreover, the Federated Defense Model is compared with the existing state-of-the-art FL models. The findings suggest that integrating FL with deep learning techniques bolsters IoT security and mitigates privacy risks and scalability challenges inherent in centralized approaches. This work contributes to the ongoing evolution of privacy protection strategies in the IoT domain, emphasizing the role of privacy-preserving methodologies in developing resilient digital ecosystems.