Abstract: In this research, we propose a DDoS attack detection and mitigation framework using machine learning for Software-Defined Networking (SDN). Rule-based or statistical models are commonly employed in existing DDoS detection methods, which fall short in terms of generalization to evolving DDoS attacks. In this paper we propose a deep learning based Dynamic Network Traffic Anomaly Detection model based on Hybrid Convolutional Neural Network and Long Short Memory approach. To ensure that the framework can withstand both known and zero-day attacks, it is trained on three heterogeneous datasets: CICIDS- 2017, UNSW-NB15, and Mininet-generated traffic. The key point is that the SDN controller serves as the central intelligence node in this architecture, receiving network flow statistics and using ML-based threat detection to apply security policies in real time. Experimental results prove that the resulting model achieves an accuracy of 99.82%, outperforms classical machine learning models and possesses strong resistance performance against adversarial attacks. The results underline the promise of using deep learning techniques for SDN environments for upcoming cyber threats.

Key words: DDoS, SDN, control plane, data plane, machine learning