A new re-sampling method for network traffic classification using SML

The way of internet traffic classification using machine learning has been a hot topic for a long time as it is independent on the packet payloads. However, the problems of classifier biasing towards the majority classes have not been solved effectively now. The uniform sampling is a popular technique to alleviate the data skew in machine learning traffic classification. But the original traffic data distribution would be destroyed by it. A new re-sampling method named tuning sampling for supervised machine learning (SML) is proposed to ease the problem of data skew in internet traffic classification. And it is compared with uniform sampling and stratified sampling methods using C4.5 classification algorithm. Our experimental results indicate that the classifier using tuning sampling gets the accuracy of minority classes are higher than the results of stratified sampling and the overall accuracy is higher than the result of uniform sampling.