A Novel LMS Method for Real-Time Network Traffic Prediction

Xinyu Yang, Ming Zeng, Rui Zhao, Yi Shi

Abstract: Real-time traffic prediction could give important information to both network efficiency and QoS guarantees. On the basis of LMS algorithm, this paper presents an improved LMS predictor – EaLMS (Error-adjusted LMS) – for fundamental traffic prediction. The main idea of EaLMS is using previous prediction errors to adjust the LMS prediction value, so that the prediction delay could be decreased. The prediction experiment based on real traffic trace has proved that for short-term traffic prediction, compared with traditional LMS predictor, EaLMS significantly reduces prediction delay, especially at traffic burst moments, and avoids the problem of augmenting prediction error at the same time.