A hybrid CNN+LSTM-based intrusion detection system for industrial IoT networks

The Internet of Things (IoT) ecosystem has proliferated based on the use of the internet and cloud-based technologies in the industrial area. IoT technology used in the industry has become a large-scale network based on the increasing amount of data and number of devices. Industrial IoT (IIoT) networks are intrinsically unprotected against cyber threats and intrusions. It is, therefore, significant to develop Intrusion Detection Systems (IDS) in order to ensure the security of the IIoT networks. Three different models were proposed to detect intrusions in the IIoT network by using deep learning architectures of Convolutional Neural Network (CNN), Long Short Term Memory (LSTM), and CNN + LSTM generated from a hybrid combination of these. In the study conducted by using the UNSW-NB15 and X-IIoTID datasets, normal and abnormal data were determined and compared with other studies in the literature following a binary and multi-class classification. The hybrid CNN + LSTM model attained the highest accuracy value for intrusion detection in both datasets among the proposed models. The proposed CNN + LSTM architecture attained an accuracy of 93.21% for binary classification and 92.9% for multi-class classification in the UNSW-NB15 dataset while the same model attained a detection accuracy of 99.84% for binary classification and 99.80% for multi-class classification in the X-IIoTID dataset. In addition, the accurate detection success of the implemented models regarding the types of attacks within the datasets was evaluated.