Towards time-varying classification based on traffic pattern

Many important network security areas, such as Intrusion Detection System and Next-Generation Firewall, leverage Traffic Classification techniques to reveal application-level protocols. Machine Learning algorithms give us the ability to identify encrypted or complicated traffic. However, classification accuracies of Machine Learning algorithms are always facing challenges and doubts in practical usage. In this paper, we propose a time-varying Logistic Regression model embedded with traffic pattern. The comparison between original Logistic Regression model and time-varying one shows an effective improvement in accuracy. We hope to exploit a new way to implement Machine Learning algorithms in network traffic analysis areas by considering the characteristics of traffic changes in time domain.