Abstract: Accurate identification of application layer traffic is important for many Internet applications, such as provision of differentiated services, quality of service(QoS) guarantee, intrusion detection, traffic monitoring, accounting, and so on. However, there is no effective method which can automatically extract application signatures. In this paper, a novel method based on frequent item mining was presented,which can automatically extract frequent set from sessions of any application protocol, reduce redundancy of the frequent set based on adaptive filtering rules, and obtain the application signatures. Identification rate and precision rate were applied to verity the extracted signatures. Experiment results show that this method is effective and the extracted signatures are subtle. Therefore it can be used to accurately identify the corresponding application.

Key words: Network application, Adaptable extraction of signatures, Frequent item mining, Session