A NOVEL APPROACH TO DETECT IOT MALWARE BY SYSTEM CALLS AND LONG SHORT-TERM MEMORY MODEL

As the Internet of Things (IoT) devices become voguish, malware detection on IoT devices is crucial today. In this paper, a novel approach to detect IoT malware based on dynamic analysis and deep learning is proposed. Our method combines an IoT-sandbox to extract system call sequences that are considered as sentences in natural language, then two Long Short-Term Memory (LSTM) model are used to classify. In our approach, a program is determined whether malware or benign by two representative values which are the results of LSTM models. Experiment results show that our proposed method outperforms other based-line machine learning models using similar system call feature in terms of accuracy, F1-Weight and the length of system call sequence. Our method uses quite short system call sequence of 150, but the highest accuracy 98.37 per cent and F1-Weight achieves 98.38 per cent. Therefore, the method can be used in early IoT malware detection solutions