Early detection of network applications using neural networks

Network traffic classification is essential in several fields such as network security and design, or providing adequate QoS. Early detection of network applications is an important feature, which enables recognition of the application online, without the need to capture and examine whole network flow. In this paper we propose a method for classification of Internet applications in the first phases of connection using sizes of first several packets only. For this purpose we used 2 types of neural network (multilayer perceptron and radial basis function network). We examine the performance of this approach for various number of packets from flow and map packet sizes to bins to decrease computational complexity.