IoT Malware Detection and Classification using Number Theory based Secure Cryptographic Deep Learning Framework

Abstract

The rapid development of Internet of Things (IoT) Networks has made the Internet of Things devices more vulnerable to high-end malwares and thus there is a demand for the precise malware detection with efficient data security. In this research, a number-theory based secure cryptographic deep-learning system is proposed for the detection and classification of IoT Malware threats. The proposed method includes the following four process of duplication removal, treatment of missing values with median, normalization of Min-Max, Autoencoder Based Feature Denoising (AEFD), Principal Component Analysis (PCA) and Transformer-based classification. Another cryptographic mechanism based on a number theory is also embedded to provide secure communications and to optimize the management of the session keys. It can be seen from the experimental results that AEFD achieves a reconstruction loss of 0.020 in only 30 epochs, which is reduced by 76.5% and the first 6 principal components still explain 89.7% of the total variance. The Transformer's training accuracy is 97.8% and its validation accuracy is 96.9%, while the proposed classifier obtains 98.0%, 97.5%, 98.2%, 97.9% and 98.4% in terms of accuracy, Precision, Recall, F1-score and ROC-AUC, respectively. Moreover, cryptographic operations are only 3.53 ms, showing that the framework is able to provide strong malware detection and secure communication but with a small computation overhead, which makes it suitable for resource-constrained IoT environments.