A bayes-updating based method for traffic classification

This paper presents a new update approach based on machine learning for accurate network traffic classification, which overcomes the innate drawback of the widely applied Naive Bayes method - accuracy and robustness drop due to changes of network flows and increases of network applications. The improvement is achieved by update the original model with new data sets sampled from the network, so as to eliminate the unfavorable influence on the output of the original model due to insufficient training data, and eventually obtain better classification results and efficiency. Theoretical analysis and experimental results show that, the proposed method enables the traffic classification model to maintain a good overall performance over time, and reduces the system overhead as well.