An efficient federated learning based defense mechanism for software defined network cyber threats through machine...

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Network Security

Background:

• Software-Defined Networking (SDN) offers flexibility and centralized control but is vulnerable to sophisticated cyber-attacks due to its architecture.
• Traditional intrusion detection systems struggle with the dynamic, large-scale nature of modern threats, leading to high false positives and slow responses.

Purpose of the Study:

• To propose an AI-based federated defense system to enhance the resilience and security of SDN environments against complex cyber-attacks.
• To improve the accuracy and speed of threat detection and response in SDN networks.

Main Methods:

• Developed an AI-based federated defense system integrating XGBoost for threat identification and LightGBM for real-time adaptive responses.
• Utilized Federated Learning to enable secure intelligence sharing among network components without compromising data confidentiality.
• Employed feature engineering on high-dimensional network traffic, log files, and system activities to enhance anomaly differentiation.

Main Results:

• Achieved a 96.3% detection rate on benchmark datasets (NSL-KDD and CICIDS2017).
• Demonstrated a 7.8% improvement in effectiveness compared to traditional Machine Learning (ML)-based intrusion detection systems.
• Significantly minimized false positives and response times.

Conclusions:

• The proposed AI-based federated defense system offers a scalable, privacy-preserving, and flexible solution for bolstering SDN security.
• The framework effectively addresses zero-day threats and is well-suited for contemporary SDN environments.