Temporal Dynamics and Spatial Content in IoT Malware detection

Abstract

Internet Of Things (IoT) has evolved as a computational platform which can facilitate large scale computing using smart devices which are connected to each other over the Internet backbone. The increasing presence of the IoT along with its processing capabilities make this environment a common target of IoT malware attacks. In recent years, many malware detection models have come up which use techniques like machine learning and deep learning. IoT malwares can be detected using either a signature-based approach or by using a behavior-based approach. Although, the signature based approach works well in the detection of known threats, it is not very effective in detecting unknown threats. Most signature based approaches require one to have the domain knowledge and the need to use various pre-processing steps before detecting the malwares. This is not suitable for real time malware detection. In this work, we build a model to tackle the malware detection problem in IoT, using the raw byte-sequences which eliminates the need for domain specific knowledge. This allows us to perform real time malware detection with lesser computational requirements. In this paper, we discuss the various challenges in creating a neural network model and propose ways to address these challenges. The experimental results show that the proposed system can classify the IoT binaries as malware or benign with an accuracy of 99.01%.