Investigating the security issues of multi-layer IoT attacks using machine learning techniques

The rapid growth of the IoT applications in smart homes, medical devices and industry 4.0 has brought huge benefits to our society and industry however faces several challenges in terms of security and privacy issues. Due to its heterogeneous architecture, lack of standard security frameworks between layers, lack of secured communication protocols and lack of resources of power, memory and cryptographic functions on edge control, current IoT devices are prone to security attacks at all three main layers: application, network, and physical layer. While there are some attacks such as Sybil, Blackhole, and Malware usually target a specific layer of IoT architecture, other attacks like DDoS and MITM can compromise the security across multiple layers. These types of attacks not only result in the loss of control over sensitive data and the theft of personal information, but they also result in financial and reputational damages. In recent years, the majority of the research has been carried out to develop robust intrusion detection technologies to safeguard the IoT from a wide range of security attacks, either through novel secure network protocols, the use of firmware or encryption techniques. It is also important to detect IoT attacks at different layers while securing it. Consequently various supervised and unsupervised machine learning (ML) techniques have been found to be effective in detecting them. There is still a lack of research for detecting multilayer attacks as well as new types of attacks in IoT devices. Hence a thorough investigation of multi-layer Intrusion Detection Systems (IDS) is provided in this article to identify a wide variety of multi-layer attacks and their behavioural patterns. The list of effective ML algorithms is also reviewed along with their training datasets. We have highlighted the challenges and opportunities for future directions of multi-layer intrusion detection research through the taxonomy of multi-layer attacks. We aim to develop a novel computational framework by investigating the similarities in the features of multi-layer attacks for training ML models to overcome existing problems.