A New DSGRU-Based Intrusion Detection Method for the Internet of Things

Abstract

The Internet of Things (IoT), a rapidly developing technology, connects entities to the Internet through information sensing devices and networks. Recently, IoT has gained widespread application in daily life and work due to its high efficiency and convenience. However, with the rapid development of IoT, the systems are intruded upon by malicious users and hackers more and more frequently. As a result, intrusion detection has attracted significant attention, and numerous schemes have been proposed that can precisely identify malicious intrusion operations. However, the existing schemes suffer from several severe challenges, such as low accuracy, high computational overhead, and poor real-time performance, in processing large-scale, high-dimensional, and temporally correlated IoT network traffic data. To address these challenges, we propose a new intrusion detection scheme for IoT in this paper. Specifically, we first improve the traditional Gate Recurrent Unit (GRU) and design a novel neural network model, namely, the Deep Supplement Gate Recurrent Unit (DSGRU). This model comprises an Original Gate Recurrent Unit (OGRU), a Decode Gate Recurrent Unit (DGRU), and a Softmax activation function. Compared with the traditional GRU, our proposed DSGRU can more efficiently extract features from IoT network traffic data and reduce the loss of features caused by nonlinear transformations during the learning process. Subsequently, we adopt DSGRU to design a novel intrusion detection scheme for IoT. We also analyze the theoretical computational complexity of the proposed scheme. Finally, we implement our proposed intrusion detection scheme and evaluate its performance based on the UNSW-NB15 and NSL-KDD datasets. The experimental results demonstrate that our proposed DSGRU-based intrusion detection scheme achieves better performance, including in terms of Accuracy, Precision, Recall, F1_score, loss value, and efficiency.