Intrusion detection model of internet of things based on deep learning

Abstract

The proliferation of Internet of Things (IoTs) technology is being seriously impeded by insecure networks and data. An effective intrusion detection model is essential for safeguarding the network and data security of IoTs. In this pa per, a hybrid parallel intrusion detection model based on deep learning (DL) called HPIDM features a three-layer parallel neural network structure. Combining stacked Long short-term memory (LSTM) neural networks with convolutional neural net work (CNN) and SK Net self-attentive mechanism in the model allows HPIDM to learn temporal and spatial features of traffic data effectively. HPIDM fuses the acquired temporal and spatial feature data and then feeds it into the CosMargin classifier for classification detection to reduce the impact of data imbalance on the 23 performance of the Intrusion Detection System (IDS). Finally, HPIDM was experimentally compared with classical intrusion detection models and the two comparative models designed in this paper, and the experimental results show that HPIDM achieves 99.87% accuracy on the ISCX-IDS 2012 dataset and 99.94% accuracy on the CICIDS 2017 dataset. In addition, it outperforms other comparable models in terms of recall, precision, false alarm rate (FAR), and F1 score, showing its feasibility and superiority.