Deep Stacking Network for Intrusion Detection

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Network security is critical, with network intrusion detection systems (NIDS) being essential.
• Traditional NIDS struggle against advanced persistent threats, necessitating improved detection mechanisms.
• The NSL-KDD dataset is widely used but presents challenges like data imbalance and feature redundancy.

Purpose of the Study:

• To address the limitations of traditional intrusion detection mechanisms.
• To propose and evaluate a Deep Stacking Network (DSN) model for enhanced network intrusion detection.
• To improve classification accuracy and overall detection performance in NIDS.

Main Methods:

• Data preprocessing techniques including under-sampling and feature selection were applied to the NSL-KDD dataset.
• A Deep Stacking Network (DSN) model was developed, integrating multiple base classifiers.
• Performance comparison of various mainstream classifiers (Decision Tree, k-NN, DNN, Random Forest) was conducted.

Main Results:

• Individual classifiers showed promising results, with Decision Tree achieving 86.1% accuracy.
• The proposed Deep Stacking Network (DSN) model achieved a higher accuracy of 86.8%.
• The DSN model demonstrated improved detection performance compared to existing NIDS research.

Conclusions:

• The Deep Stacking Network (DSN) model offers a significant improvement in network intrusion detection accuracy.
• Data preprocessing techniques are crucial for handling challenges in intrusion detection datasets.
• The proposed model provides a more powerful and effective solution for network security against advanced threats.