Short-Term Prediction of SDN Network Traffic Based on CEEMDAN and Mixed Kernel Least Squares Support Vector Machine

Abstract

Nowadays, effective prediction of Software Defined Network (SDN) traffic is one of the important ways to improve network service quality and secure network services. In order to improve the prediction accuracy of SDN network traffic, this paper proposes a combined prediction model combining decomposition algorithm, hybrid kernel least squares support vector machine, and optimization algorithm, considering the characteristics of network traffic such as nonlinearity and non-smoothness. Firstly, the original data series is decomposed into a series of Intrinsic Mode Function (IMF) using the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to smooth the series. Then, the fuzzy C-mean algorithm (FCM) is used to classify each component into three classes based on its amplitude-frequency characteristics. Next, based on the different characteristics of the kernel functions, least-squares support vector machine (LSSVM) prediction models with corresponding kernel functions are constructed for each class of components, and the parameters of each model are optimized using the artificial bee colony algorithm (ABC). Finally, the prediction results of each component are cumulatively reconstructed to obtain the final prediction results. Through experimental comparison, the prediction accuracy of the proposed model (CEEMDAN-FCM-LSSVM-ABC) is better than that of CEEMDAN-LSSVM, EMD-LSSVM, LSSVM, and other models.