An efficient intrusion detection system for IoT security using CNN decision forest.

Area of Science:

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

Background:

• The widespread adoption of IoT devices has increased vulnerability to cyberattacks.
• Intrusion Detection Systems (IDS) are critical for securing IoT ecosystems.
• Existing IDS methods require enhancement for improved accuracy and efficiency.

Purpose of the Study:

• To introduce a novel Deep Neural Decision Forest-based IDS (DNDF-IDS) for enhanced network anomaly detection.
• To improve the accuracy and efficiency of intrusion detection in IoT environments.
• To evaluate the performance of DNDF-IDS using various feature selection techniques.

Main Methods:

• Developed a novel DNDF-IDS integrating decision forests with neural networks.
• Applied four feature selection methods: PCA, LASSO, SelectKBest, and RFFI.
• Evaluated the model on three benchmark datasets: NSL-KDD, CICIDS2017, and UNSW-NB15.

Main Results:

• Achieved high accuracy (ACC) ranging from 94.09% to 98.84% across datasets.
• Demonstrated a prediction time of 0.1 ms per record.
• Outperformed or matched recent Random Forest and CNN-based models, especially with top 10 features.

Conclusions:

• The DNDF-IDS offers a highly accurate and efficient solution for IoT security.
• The model effectively identifies network anomalies using a minimal set of features.
• This approach significantly improves computational resource utilization in intrusion detection.