IoT Intrusion Detection Taxonomy, Reference Architecture, and Analyses.

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Internet of Things

Background:

• Deep learning (DL) offers advanced capabilities for intrusion detection systems (IDSs).
• The Internet of Things (IoT) ecosystem faces unique security challenges requiring specialized IDSs.
• Existing IDSs often lack comprehensive evaluation in diverse IoT environments.

Purpose of the Study:

• To survey and analyze deep learning approaches for IoT intrusion detection.
• To identify gaps, weaknesses, and propose a reference architecture for DL-based IDSs in IoT.
• To provide a comparative study of anomaly-based IDSs utilizing DL techniques.

Main Methods:

• Reviewed supervised, unsupervised, and hybrid DL methods for anomaly-based IDSs.
• Evaluated feature extraction, classification, prediction, and regression implementations.
• Assessed performance metrics, detection rates, and efficiency of various DL algorithms.
• Compared Logistic Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), and Artificial Neural Network (ANN) using Receiver Operating Characteristic (ROC) curves.

Main Results:

• Deep learning approaches show promising outcomes for detecting various attack classes in IoT.
• Anomaly-based IDSs using DL techniques are effective in IoT environments.
• The study provides insights into the performance of different DL models on benchmark datasets like Bot-IoT and IoTID20.

Conclusions:

• Deep learning is a viable and effective technology for enhancing IoT security.
• Further research is needed to address identified gaps and refine reference architectures for DL-based IoT IDSs.
• The comparative analysis offers valuable guidance for selecting and implementing DL models for IoT intrusion detection.