Deep Stacking Network for Intrusion Detection.

Area of Science:

• Computer Science
• Cybersecurity

Background:

• Network intrusion prevention is critical for robust network security.
• Traditional intrusion detection mechanisms struggle with advanced cyber threats.
• Existing intrusion detection systems require enhancement to counter sophisticated attacks.

Purpose of the Study:

• To analyze the NSL-KDD dataset for network intrusion detection.
• To address data imbalance and feature redundancy issues in intrusion detection datasets.
• To propose and evaluate a novel Deep Stacking Network model for improved intrusion detection accuracy.

Main Methods:

• Data preprocessing techniques including under-sampling and feature selection were applied to the NSL-KDD dataset.
• A Deep Stacking Network model was developed by integrating multiple base classifiers.
• Performance comparison of various mainstream classifiers like decision trees, k-nearest neighbors, deep neural networks, and random forests was conducted.

Main Results:

• Individual classifiers like decision trees achieved 86.1% accuracy.
• The proposed Deep Stacking Network model, fusing four classifiers, reached an accuracy of 86.8%.
• The Deep Stacking Network demonstrated superior performance compared to existing intrusion detection systems in research.

Conclusions:

• The developed Deep Stacking Network model significantly enhances network intrusion detection performance.
• Data preprocessing techniques effectively managed dataset imbalances and redundancy.
• The study highlights the potential of ensemble methods for building powerful intrusion detection systems.