Early recognition of Internet service flow

Early service flow recognition is required in both network management and intrusion detection systems. However, there exists some deficiency in existing machine learning methods when used in time or accuracy critical conditions. We put forward an early recognition method of Internet service flow and verify it by multiple experiment sets. As simulation results show, based on the packet lengths and the packet intervals of the first N packets, the method, utilizing improved k nearest neighbors (KNN) algorithm, can effectively identify the unknown service flow in the network.