Comparative analysis of deep learning and traditional methods for IoT botnet detection using a multi-model framework...

Area of Science:

• Cybersecurity
• Machine Learning
• Internet of Things (IoT)

Background:

• The rapid expansion of Internet of Things (IoT) devices has introduced significant cybersecurity vulnerabilities.
• Botnets pose a critical threat to network infrastructure, exploiting these vulnerabilities.

Purpose of the Study:

• To develop a robust and accurate detection system for IoT botnets.
• To enhance cybersecurity defenses against emerging threats in IoT environments.

Main Methods:

• A novel ensemble framework integrating Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), Random Forest (RF), and Logistic Regression (LR) using weighted soft-voting.
• Implementation of Quantile Uniform transformation for feature skewness reduction and a multi-layered feature selection method.
• Evaluation of the framework on BOT-IOT, CICIOT2023, and IOT23 datasets.

Main Results:

• Achieved 100% accuracy on the BOT-IOT dataset.
• Demonstrated 99.2% accuracy on the CICIOT2023 dataset and 91.5% on the IOT23 dataset.
• Outperformed existing state-of-the-art models by up to 6.2%.

Conclusions:

• The proposed ensemble framework offers a scalable and high-performance solution for IoT botnet detection.
• The model is adaptable to diverse network scenarios, providing practical optimizations for real-world deployment.
• This research significantly advances IoT security by enabling more effective threat detection.