SmartDefender: Semi-Supervised Intrusion Detection System for Multi-Attack IIoT Environment

Abstract

Due to the scale of the Industrial Internet of Things (IIoT), the extensive use of varied types of devices, and continuous data exchange, these types of networks are increasingly vulnerable to cyberattacks. Traditional intrusion detection methods do not work well, as they cannot detect complex and evolving attack patterns in these environments. Therefore, this project presents an IDS that takes a web-based approach and utilizes deep learning techniques for intrusion detection by integrating CNN-GRU model training on the UNSW-NB15 dataset to systematically identify meaningful spatial and temporal features of network traffic data. The IDS uses feature scaling and label encoding to ensure the consistency of the datasets during preprocessing and improve the accuracy of classifications. Additionally, a semi-supervised machine learning classifier has been constructed to detect normal behaviour and multiple types of attacks. The complete system was developed within a Streamlit-based web application. Users can upload their network traffic files in either CSV or Parquet format, which will then be processed by the trained IDS machine learning based model(s) and displayed in either a table or graphical format, providing a fast, trusted, and simple way to perform intrusion detection for any IIoT environment.