Industrial Internet of Things Cyber Threats Detection Through Deep Feature Learning and Stacked Sparse Autoencoder Based Classification

Abstract

In recent times, the industrial system has integrated with industrial Internet of Things (IoT) applications to enable the ease of production process and ensure worker safety. Security is a major concern in the industrial Internet of Things (IIoT) environment owing to the distributed nature of architecture and dynamic traffic flows. Generally, the cyber-attack detection model is classified as misuse and anomaly detection. The misuse detection method is employed based on the concept of signature matching, and the anomaly method is based on the detection of known and unknown attacks. Present security models have realized the issue of over-fitting, low classification accuracy, and a high false positive rate when given a massive volume of network traffic data. The proposed work focused on “IIoT cyber-attack detection using lightweight hybrid deep learning algorithm” to identify intrusion. At first, the data imbalance problem is resolved through the Euclidean-based synthetic minority oversampling technique (EbSmoT) to prevent the model from becoming biased toward one class. Then, the Information Gain and Fisher score-based technique (IG-FST) is employed to eliminate redundant features and avoid overfitting problems during training. Moreover, the Bi-LSTM ResNet-based convolutional autoencoder (BR-CAE) is executed to obtain higher-level feature representation. Finally, a Stacked Sparse autoencoder-based Particle Swarm Probabilistic Neural Network (SAE-PSPNN) is used for attack detection and classification. The performance of the proposed method can be evaluated using several performance metrics through two different datasets, such as the UNSW-NB15 dataset and the ToN_IoT dataset. The proposed framework achieved an accuracy of 99.86% on the ToN_IoT dataset and 99.62% on the UNSW-NB15 dataset.