Ensemble learning model for P2P traffic identification

P2P traffic identification is an important issue of internet traffic analysis, and machine learning is a viable approach to address it. However, compared to ensemble learning methods, traditional methods and simple machine learning methods appear to be slightly limited in improving performance. In this paper, Random Forests and feature weighted Naive Bayes was integrated to P2P traffic identification. Scores were calculated for each category in the model while the process of prediction. Then, weighted majority voting was used to get the final output. Experiments were conducted to verify the effectiveness and stability of the integrated model, which implements in the programming mode of MapReduce. Results have shown that the model achieved a better overall performance and may provides an alternative way to solve P2P traffic identification problem.