Research on combination prediction of Web traffic based on wavelets

For improving the prediction accuracy of Web traffic, a novel combination prediction method was proposed based on the integration of the wavelet, neural network and the auto-regression (AR). Two correlative traffic series, history series and similar values series, were distilled from the web traffic data. The stationary similar values series was predicted by AR model. The nonlinear and non-stationary history series were decomposed and then reconstructed into several branches by wavelet. These branches were predicted by neural networks or AR models respectively according to their different features. The predicted results of the two series were combined into the final predicted value. The results show that the combination prediction can take advantage of diverse correlative data relationships. The wavelet analysis can decompose history series into several time serials that have simpler frequency components and are easier to be forecasted. So the method has better predictive precision compared with traditional prediction approaches.