IoT Malware Detection Technique Based on Deep Learning and Natural Language Processing

The Internet of Things (IoT) has gained significant popularity and attention as a research topic due to its widespread use and associated concerns. Previous security measures, which focused on specific threats using limited and outdated datasets, need to be revised to address the growing size and diversity of IoT applications. Consequently, there is a pressing need to prioritize security in IoT systems. Various deep learning (DL)-based algorithms have been developed to address this challenge for IoT attack detection. In this study, we proposed two deep learning models for malware classification integrated with natural language processing (NLP) techniques, such as padding and word embedding. The first model utilizes a Recurrent Neural Network (RNN), and the second employs a one-dimensional Convolutional Neural Network (CNN). Both proposed systems were trained and evaluated on 3971 opcode instruction files representing attacks on nine CPU architectures. The proposed methods were evaluated based on accuracy, F1-score, recall, and precision and were compared with the latest approaches. The experimental results demonstrate the effectiveness of the proposed models in accurately identifying and categorizing IoT attacks. The CNN classifier achieved an accuracy of 96.7%, while the RNN classifier achieved an impressive accuracy of 99.3%.