Transfer Learning for Internet of Things Malware Analysis

Abstract

Internet of Things (IoT) environments are characterized by devices that have heterogeneous applications, diverse underlying technologies and most of them are constrained in resources such as low memory, and weak security mechanisms. The synthesis of malware that menaces the Internet of Things environments is an open and evolving research problem. This study proposes use of images and transfer learning to analyze IoT malware. The malware files are converted to three channel images to cull architecture and platform of analysis dependence. The preprocessed malware images are split to training and testing set. The deep neural network is based on pretrained VGG19 Model and adapts the last layer to discriminate the malware images into their malware families. The experimental results on the dataset adopted from IoTPoT dataset shows that our approach can be used effectively to classify malware into families as it attains 89.23% overall accuracy and F measure of 91.3%.