Robust prediction of network traffic using Quantile Regression Models

Reliable network traffic prediction is essential for efficient resource management schemes. Based on the quantile regression, we propose a robust prediction procedure which is resistent to outliers. For long-term predictions, the predicting intervals have a coverage probability that is very close to the pre-assigned nominal level. The detailed distributional information of the estimated quantities can be efficiently characterized by using different quantiles. The performance of the prediction is tested on a large telecommunication network traffic data. The results indicate that the proposed quantile regression provide relative accurate prediction and is not sensitive to outliers.