An efficient federated learning based defense mechanism for software defined network cyber threats through machine learning models
Rashid Amin, Antonio Costanzo, Lial Raja Alzabin, Aqsa Aqdus, Syafiq Fauzi Kamarulzaman, Asma Hassan Alshehri · Scientific Reports · 2025
Software-Defined Networking (SDN) is flexible, provides centralized control, and is vendor-independent; however, due to its centralized architecture, it is extremely susceptible to complex cyber-attacks. The classical means of intrusion detection proves to be unable to cope with dynamic and large-scales threats, which prove to be high-false positive as well as slow responses. This work proposes an AI-based federated defense system, which incorporates XGBoost to identify threats correctly, LightGBM to provide adaptive responses in real-time, and Federated Learning to share intelligence among each other without jeopardizing data confidentiality. Based on high-dimensional network traffic, log files, system activities, the framework uses feature engineering to improve anomaly differentiation and boost SDN resilience. Extensive testing on benchmark datasets (NSL-KDD and CICIDS2017) shows that the suggested solution has a 96.3% detection rate, which minimizes false positives and is 7.8% more effective than traditional ML-based intrusion detection systems. The framework can be easily scaled, it saves privacy, and is flexible to new zero-day threats, fitting the current SDN-based environments.