A Traffic Classification Algorithm Based on Neural Network

A number of network activities can benefit from accurate traffic classification and identification. However, the current classification methods, either port-based or payload-based, are becoming ineffective as many P2P applications use dynamic port numbers, masquerading techniques, and encryption to avoid detection. This paper discusses and explores a new method that is based on supervised machine learning mechanisms, which eliminate the inherent limitations of port-based or payload-based methods. A series of experiments show that, combined with a fast correlation-based feature selection filter, better performance and more accurate identification results can be obtained using the neural network method. Due to all the favorable features and satisfactory performance, the proposed methods are promising for internet traffic classification.