Evaluation of dimensionality reduction on network traffic pattern recognition.

The traffic flow on a network at time t is defined by a vector Vt of dimensionality L comprising the Link volumes on the network. This paper examines four techniques for reducing the dimensionality of V-t and evaluates these techniques in terms of the resulting effectiveness of a network traffic pattern matching scheme. The results indicate that a significant reduction in dimensionality can be attained with little sacrifice in reliability.