A new P2P traffic identification methodology based on flow statistics
HuiLin Chu, YI Hong-bo, Xingming Zhang · 2011
Nowadays P2P traffic consumes a great amount of network bandwidth which brings many difficulties to network management. In order to accurately identify P2P traffic, this paper proposes a methodology based on flow statistics. At first it quickly eliminates those flow features irrelevant to class by the ReliefF algorithm, then from the rest features it uses a wrapper method combined genetic algorithm with support vector machine to select flow features and optimize the parameters of support vector machine model, and finally it outputs the best flow feature set and the optimized support vector machine model. The experimental results indicate that this methodology can achieve improved accuracy with fewer flow features.