A novel method of network burst traffic real-time prediction based on decomposition

Network traffic burst becomes a threat to network security. In this paper, a decomposition based method is presented for network burst traffic real-time prediction, in which, by passing smoothing filter, network traffic is decomposed into smooth low frequency traffic and high frequency traffic to make prediction respectively, and then a superposition result of the predictions is yielded. Based on LMS algorithm, an improvement of LMS predictor by adjusting prediction according to prediction errors (EaLMS, Error-adjusted LMS) is proposed to process the low frequency traffic, and a simple method of linear combination is presented to predict the high frequency traffic. The experiment results using real network traffic data shows, compared with traditional LMS, the prediction method based on decomposition obviously shorted the prediction delay and reduced the prediction error during traffic burst, while it also improves the global prediction.