An Effective Anomaly Detection Model Combined Feature Selection with Improved CNN in SDN

The frequent intrusion of anomalous traffic has consistently remained a critical concern in the domain of network security. Software-Defined Networking (SDN) is an innovative architecture that outperforms traditional architectures. However, the decoupled nature of its central control layer and underlying data devices makes it vulnerable to anomalous intrusions. To address the shortcomings of existing anomaly detection methods in terms of incomplete feature selection and insufficient detection efficiency, we propose an improved Convolutional Neural Network (CNN) based anomaly traffic detection model for SDN architecture. An improved swarm intelligence optimization algorithm (OGSSA) is utilized to filter the optimal subset of features, followed by a CNN that references the attention module (ECA- CNN) as the classifier. We perform experimental simulations using the InSDN dataset and compare it with a variety of traditional algorithms and deep neural networks. The findings indicate that the feature mixture algorithm and CNN detection model we designed are appropriate and effective in identifying normal and malicious network activities.