Applications of nonlinear prediction methods to the Internet traffic

Abstract

In this paper, nonlinear time series prediction methods are applied to the Internet traffic. First, the generic local linear approximation method, the radial basis function networks and the support vector machines are applied to prediction of chaotic time series in order to evaluate these methods. Then, a sample version of the local linear approximation method is selected because it is easy to apply and has high predictability. It is applied to various Internet traffic data sampled at different times. As a result, cross correlation coefficients between the actual traffic time series and predicted time series was larger than 0.9 on some in those sampled data sets. Moreover, the effectiveness of applications of nonlinear time series prediction methods to the traffic data is confirmed by the method of surrogate data.