Many research efforts propose using flow-level features (e.g., packet sizes and inter-arrival times) in tandem with machine learning algorithms to solve traffic classification. In this respect, one of the most significant drawbacks of state-of-the-art solutions is that they are modeled on data flows assumed to be completed by natural expiration, eventually causing a massive variability in the prediction times.

Network traffic subflow that reflects the current state of only a fraction of the total traffic flow appears a more appropriate approach to classification as it is suited for operation in real-world applications by nature. However, its better understanding requires in-depth examination in a variety of use cases.

The task of the student is to examine network traffic subflow characteristics for traffic classification based on machine learning. In the course of the work, the student will learn about network traffic subflow measurement techniques, how to analyze and use the measured data, especially with respect to flow state transition, and the most popular methods for machine learning-based classification. Building on the knowledge acquired, the student will provide solutions for subflow state transition modeling using machine learning.