Enhanced botnet detection in IoT networks using zebra optimization and dual-channel GAN classification

Area of Science:

• Computer Science
• Cybersecurity
• Network Security

Background:

• The Internet of Things (IoT) is expanding across critical sectors like healthcare and smart cities.
• IoT devices face significant security risks due to limited processing power and inadequate security protocols.
• Existing antimalware solutions struggle to address the evolving landscape of IoT threats.

Purpose of the Study:

• To develop an advanced, feature selection-based classification model for robust IoT malware detection.
• To enhance the accuracy and reliability of identifying anomalous activities, such as botnet intrusions, within IoT networks.

Main Methods:

• A preprocessing stage involving data smoothing and consistency improvement was implemented.
• The Zebra Optimization Algorithm (ZOA) was employed for effective feature selection and dimensionality reduction.
• A Dual-channel Graph Attention Network (DGAN), incorporating Node and Semantic Attention Networks, was utilized for classification.
• The Sooty Tern Optimization Algorithm (STOA) was applied for hyperparameter tuning to optimize model performance.

Main Results:

• The proposed STOA-DGAN model achieved a high classification accuracy of 99.87% for botnet activity.
• The model demonstrated superior robustness and reliability compared to existing cybersecurity approaches.
• Integration of structural and semantic data significantly improved detection capabilities.

Conclusions:

• The STOA-DGAN model presents a highly effective solution for detecting botnet activities in IoT environments.
• The feature selection and advanced network architecture contribute to enhanced security against sophisticated cyber threats.
• This research offers a reliable method for securing critical IoT infrastructure and applications.