Recognition of internet traffic using ML and a statistical method

Abstract

Researchers have been exploring ways for identifying Internet activity that is not dependent on 'well-known' TCP or UDP port numbers or packet payload interpretation. Other methods use statistical patterns in the traffic's externally visible characteristics to categorize it (such as typical packet lengths and inter-arrival times). Classifying Internet traffic flows into clusters with similar statistical features is the fundamental objective. To cope with traffic patterns, huge datasets, and multidimensional domains of flow and packet characteristics, machine learning (ML) approaches were introduced in this sector.