Smart Hybrid Intrusion Detection for IoT Networks Using Machine Learning and Neural Networks

Abstract

With the fast growth of Internet of Things (IoT), there is an increasing security impact, e.g., IoT environments has become the advanced target of complex cyber-attack like Distributed Denial of Service (DDoS) attack. Most conventional Intrusion Detection Systems (IDSs) fail to scale with the heterogeneity, scale, and dynamism of IoT networks. To overcome these challenges, in this paper, we develop a smart hybrid intrusion detection framework that can effectively utilize the power of Machine Learning (ML) and Neural Network (NN) models, namely: cascades backpropagation neural network (CPBNN) and convolutional neural network (CNN) to improving accuracy and adaptability of detection in IoT environments. The proposed system employs a two-layer detection structure where the CPBNN model is used to detect abnormal patterns in the network packet level and identify the abnormal behaviors of packet sequences, whereas the CNN model performs deep feature extraction and classification to predict the abnormality as well as the nature of the abnormality. We develop this hybrid architecture to work well under IoT level large-scale deployments without incurring software overhead. The system has been evaluated on the KDDTest-21 benchmark dataset with the usual metrics, including accuracy, precision, recall and F1-score. Experiments results prove the CNN’s performance with 90% and CPBNN with 82% which valid the efficiency of the proposed method in detecting IoT related security threats. These results suggest the ability of intelligent hybrid IDS systems not only to defend in a pro-active manner using ML and neural networks in synergy, but also to use the adversary's strength against itself.