Design Guidelines for Machine Learning-based Cybersecurity in Internet of Things

Abstract

Cybersecurity is one of the building blocks in need of increasing attention in Internet of things (IoT) applications. IoT has become a popular target for attackers seeking sensitive and personal user data, computing infrastructure for massive attacks, or aimed at compromising critical applications. Worryingly, the industrial race toward the forefront of IoT software and device development has led to increased market penetration of vulnerable IoT devices and applications. Nevertheless, traditional cybersecurity solutions designed for personal computers often rely on heavy computation and high communication overhead, and therefore are prohibitive for IoT, given the explosive number of IoT devices, their resource-con-strained nature, and their heterogeneity. Hence, innovative solutions must be designed for securing IoT applications, while considering the peculiar characteristics of IoT devices and networks. In this article, we discuss the motivations and challenges of using machine learning (ML) models for the design of cybersecurity solutions for IoT. More specifically, we tackle the challenge of designing ML-based solutions and provide guidelines for ML-based physical layer solutions aimed at securing IoT. We propose a device-oriented and network-oriented classification and investigate recent works that designed ML-based solutions, considering IoT physical layer features, to secure IoT applications. The proposed classification helps engineers and practitioners starting in this area to better identify and understand the challenges, requirements, and up-to-date common design principles for securing IoT devices and networks considering physical layer features. Finally, we shed light on some future research directions that need further investigation.