BiLSTM-CNN Hybrid Intrusion Detection System for IoT Application

Abstract

Intrusions in computer networks have increased significantly in recent times and network security mechanisms are not being developed at the same pace at which intrusion attacks are evolving. Therefore, a need has arisen to improve intrusion detection systems (IDS) to make network secure. This research focuses on anomaly-based IDS for security assaults. In this research, deep learning techniques such as Bi-directional Long Short-Term Memory (Bi-LSTM) and Convolutional Neural Networks (CNN) are implemented and subsequently used to design a novel BiLSTM-CNN hybrid IDS for the Internet of Things (IoT). The hybrid intrusion detection system model is created by utilizing the advantages of both the BiLSTM and the CNN's ability to extract temporal and spatial features respectively. The research uses the UNSW-NB 15 dataset for proposed deep learning IDS for IoT networks. The dataset has been split into training and testing data for classifying traffic into normal or attack classes. The models are run on GPU and CPU to illustrate their efficacy and match real-world IoT network communication behavior. The BiLSTM, CNN, and hybrid BiLSTM-CNN models are assessed on various aspects like Precision, Sensitivity, F1-Score, Miscalculation Rate, False Positive Rate, False Negative Rate, and Matthews Correlation Coefficient to evaluate the model’s robustness. The findings revealed that the hybrid model surpassed the BiLSTM and CNN models in all aspects. Additionally, the proposed model is compared with the cutting-edge existing approaches in terms of different performance metrics and proved to be better than state-of-the-art models.