Internet traffic prediction plays a fundamental role in network design, management, control, and optimization. The self-similar and non-linear nature of network traffic makes highly accurate prediction difficult. In this paper, a boosting-based framework is proposed for self-similar and non-linear traffic prediction by considering it as a classical regression problem. The framework is based on Ada-Boost on the whole. It adopts Principle Component Analysis as an optional step to take advantage of self-similar nature of traffic while avoiding the disadvantage of self-similarity. Feed-forward neural network is used as the basic regressor to capture the non-linear relationship within the traffic. Experimental results on real network traffic validate the effectiveness of the proposed framework.
|Original language||English (US)|
|Number of pages||6|
|Journal||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|State||Published - Dec 1 2004|
ASJC Scopus subject areas
- Theoretical Computer Science
- Computer Science(all)