Machine learning based internet traffic recognition with statistical approach

The researchers have started looking for Internet traffic recognition techniques that are independent of `well known´ TCP or UDP port numbers, or interpreting the contents of packet payloads. Newer approaches classify traffic by recognizing statistical patterns in externally observable attributes of the traffic (such as typical packet lengths and inter-arrival times). The main goal is to cluster or classify the Internet traffic flows into groups that have identical statistical properties. The need to deal with Traffic patterns, large datasets and Multidimensional spaces of flow and packet attributes is one of the reasons for the introduction of Machine Learning (ML) techniques in this field. ML techniques are subset of Artificial Intelligence used for traffic recognition. Further, there are four types of Machine Learning, i.e. Classification (Supervised learning), clustering (Un-Supervised learning), Numeric prediction and Association. In this research paper IP traffic recognition through classification process is implemented. Different researchers are calling this process as IP traffic Recognition, IP traffic Identification, and sometimes IP traffic classification. Here Real time internet traffic has been captured using packet capturing tool and datasets has been developed. Also few standard datasets have been used in this research work. Then using standard attribute selection algorithms, a reduced statistical feature dataset has been developed. After that, Six ML algorithms AdaboostM1, C4.5, Random Forest tree, MLP, RBF and SVM with Polykernel function classifiers are used for IP traffic classification. This implementation and analysis shows that Tree based algorithms are effective ML techniques for Internet traffic classification with accuracy up to of 99.7616 %.