Safeguarding IoT networks against DDoS attacks using deep learning based zero trust network access

• Published in: Electronics letters 2024-11, Vol.60 (21), p.n/a
• Main Author: Khan, Murad
• Format: Article
• Language: English
• Subjects: network analysis; network topology; security of data
• ISSN: 0013-5194; 1350-911X
• DOI: 10.1049/ell2.70075

Here, a deep learning‐based zero trust network access (DL‐ZTNA) system to enhance the security of the Message Queuing Telemetry Transport (MQTT) protocol within Internet of Things (IoT) applications was proposed. Combining multi‐head convolutional neural networks and attention‐based bi‐directional long short‐term memory networks with ZTNA provides real‐time security analysis of device behaviour. This behaviour‐based approach ensures that only authorized devices can access network resources and continuously monitors for potential threats like distributed denial of service (DDoS) attacks. The proposed DL‐ZTNA system revokes device access when a threat is detected and prevents further malicious activities. Evaluation in a testbed environment showed improvements in CPU usage efficiency, throughput, and attack detection probability compared to traditional methods. This highlights the system's effectiveness in securing MQTT‐based IoT networks against DDoS attacks while maintaining high performance, showcasing the potential of integrating deep learning techniques into ZTNA system for addressing security challenges in IoT environments. The proposed deep learning‐based zero trust network access (DL‐ZTNA) system secures Message Queuing Telemetry Transport‐based Internet of Things networks by integrating multi‐head convolutional neural networks with attention‐based bi‐directional long short‐term memory networks for real‐time security analysis. This system effectively monitors and restricts unauthorized device access, particularly under distributed denial of service attacks, showing improved efficiency in CPU usage, throughput, and detection accuracy compared to traditional security methods.