A Two-Level Hybrid CNN Model for IoT Network Attack Identification

Due to the fast and significant evolution of Cyber-Physical Systems (CPS), security has become a common problem for all the users of CSP. People use the modern Internet of Things (IoT) networks in healthcare, agriculture, smart cities, manufacturing, power grid, etc. As the number of IoT applications and users increases, the number of vulnerabilities also increases parallelly. Many non-profit or government organizations have incurred significant losses due to vulnerable attacks on IoT networks. So, it is important to identify anomalies in a network at the earliest before making a significant impact on the network. Since IoT networks generate massive network traffic, traditional Machine Learning (ML) techniques require considerable time and hand-crafted features for anomaly detection. This paper presents a two-level hybrid Deep Learning (DL) model using 1D Convolution Neural Network (1D-CNN). The level 1 model classifies the network flow as normal or attacks, while the level 2 model identifies the type of attacks. We evaluated the proposed model on the UNSW-NB15 dataset and analyzed its performance in terms of accuracy, precision, recall, and F score. From the experimental analysis, we observed that the proposed model yields 94.51 and 80.54% detection accuracy on level 1 and level 2, respectively.