Resilient and Deep Network for Internet of Things (IoT) Malware Detection

Abstract

Nowadays, digital devices and the internet make our life remarkably easy since a massive number of daily activities can be carried out simply through the internet. Internet of Things (IoT) devices are increasingly employed in diverse industries with a wide range of purposes such as sensing or collecting environmental data. The development of IoT brings many opportunities but also many security challenges. Recently, the presence of IoT in a numerous number of applications and their improving computing and processing abilities make them a vulnerable attack target. Hence, developing a method that is capable of proactively detect and prevent malware in IoT is a perpetual demand. In the recent years, machine learning techniques have been applied in the field of malware detection and achieved acceptable results, however; these approaches have inherently a challenging step called feature extraction. Therefore, we need a method that has the ability to automatically extract features which is significantly time-consuming and error-prone process. The introduction of deep learning, a new area of artificial intelligence, helps the malware detection by automating the feature extraction due to its multi-layer training. This paper proposes a novel architecture of Convolutional Neural Network (CNN) that utilizes raw bytes as input and eliminates the need to extract high-level features manually. In addition, we benefit from the reputed embedding techniques to generate numerical vectors of bytes since deep networks only accept numerical vectors as input. Our results indicate that the proposed approach can achieve high detection rate of malware among IoT devices, outperforming traditional machine learning based methods which reveals the merit of deep learning techniques in IoT malware detection.