Machine Learning-Based DDoS Detection Using Variational Mode Decomposition

With the rapid advancement of informatization and digitization, Distributed Denial-of-Service (DDoS) attacks have exhibited escalating scales and frequencies. The high-dimensional, dynamic, and non-stationary characteristics of network traffic data pose significant challenges to traditional feature optimization methods. This study investigates a strategic framework integrating Variational Mode Decomposition (VMD) with Random Forest Feature Importance (RFFI) and Pearson correlation coefficients for DDoS attack detection, aiming to explore its potential in optimizing network intrusion datasets. Experimental evaluations using common machine learning models demonstrate that the VMD-based framework enhances detection accuracy, achieving approximately 7% performance gain across optimal model configurations. Notably, the K-Nearest Neighbors (KNN) method attained the highest detection accuracy of 99.55%. These findings preliminarily validate the effectiveness of the VMD-based framework in processing non-stationary network data.