ROAST-IoT: A Novel Range-Optimized Attention Convolutional Scattered Technique for Intrusion Detection in IoT...

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• The Internet of Things (IoT) presents significant security challenges due to system complexity and data volume.
• Existing Intrusion Detection Systems (IDSs) for IoT often suffer from inadequate detection, high latency, and prolonged processing times.
• Machine learning (ML) is widely adopted for IoT IDSs, but performance limitations persist.

Purpose of the Study:

• To propose a novel and efficient intrusion detection system for IoT networks.
• To address the limitations of existing IoT security frameworks, specifically latency and processing time.
• To enhance the accuracy and reliability of intrusion detection in IoT environments.

Main Methods:

• Development of the Range-Optimized Attention Convolutional Scattered technique (ROAST-IoT).
• Utilized Scattered Range Feature Selection (SRFS) for crucial feature extraction.
• Employed an Attention-based Convolutional Feed-Forward Network (ACFN) for intrusion classification.
• Optimized the classifier's loss function using the Modified Dingo Optimization (MDO) algorithm.

Main Results:

• ROAST-IoT demonstrated superior performance across multiple benchmark datasets (ToN-IoT, IoT-23, UNSW-NB 15, Edge-IIoT).
• Achieved high accuracy rates, including 99.15% on IoT-23, 99.78% on ToN-IoT, 99.88% on UNSW-NB 15, and 99.45% on Edge-IIoT.
• Attained an average AUC-ROC of 0.998, indicating excellent discrimination between normal and attack traffic.

Conclusions:

• ROAST-IoT effectively and reliably detects intrusion attacks against IoT systems.
• The proposed technique significantly outperforms existing state-of-the-art intrusion detection systems.
• ROAST-IoT offers a promising solution for enhancing the security posture of complex IoT networks.