Hybrid Machine Learning and Deep Learning Model for Efficient DDoS Attack Detection in Software-Defined Networks

Network infrastructure facing Distributed Denial of Service (DDoS) attacks represent a severe threat mainly to Software-Defined Networks (SDN) thanks to their centralized control vulnerability. DDoS detection uses traditional Machine Learning (ML) and Deep Learning (DL) approaches, yet they struggle because of high computational expenses or inadequate generalization power. This paper develops a mixed ML-DL framework that unites ML feature extraction with deep learning classification to optimize DDoS traffic detection performance. The most important features extracted from the DDoS SDN dataset result from applying Random Forest (RF) feature importance and Principal Component Analysis (PCA) and Chi-Square feature selection, effectively reducing data dimensionality. The chosen features form the input for a CNN-LSTM hybrid model, which analyzes spatial dependencies through CNN and learns temporal network patterns using LSTM. The proposed model delivers exceptional accuracy levels. At the same time, it operates efficiently, which results in higher performance than standard ML and DL techniques. This approach leads to quick training time alongside reliable detection accuracy, establishing it as a practical solution for real-time DDoS mitigation within SDN systems.