Amidst the rapid advancements in informatization and digitization, cybersecurity has become an increasingly critical issue. The rise of sophisticated attacks, such as Distributed Denial of Service (DDoS) attacks, has led to a dramatic increase in the scale and frequency of network disruptions. However, network traffic data often exhibits complex characteristics, including high dimensionality, dynamic behavior, and non-stationarity, which pose significant challenges for traditional feature optimization methods. To address these challenges, this study introduces the Variational Mode Decomposition(VMD) method into the domain of DDoS attack detection, aiming to explore its potential for optimizing network intrusion detection datasets. The effectiveness of this approach in improving DDoS attack detection performance is validated using models such as Logistic Regression (LR), Support Vector Machine (SVM), Naive Bayes (NB), and K-Nearest Neighbors (KNN). Experimental results demonstrate that the VMD-based optimization strategy significantly enhances detection accuracy, achieving performance improvements of over 7% across all models, with the KNN method achieving the best detection performance of 99.55%. These highlight the potential of VMD to effectively handle non-stationary network intrusion datasets.