Computer Network Traffic Classification Using Machine Learning Technique
- Details
- Category: IT
- Published: Monday, 15 February 2016 02:00
- Written by Nosaiba Hamdan Abu-Samhadanh
- Hits: 1335
Computer Network Traffic Classification Using
Machine Learning Technique
Nosaiba Abu-Samhadanh
Mutah University, 2015
In recent years, the uses of the Internet has increased and been extensively developed. Many modern applications have evolved to facilitate the process of social communication. Also the traffic classification process has appeared as a science in itself on the Internet nowadays.
In this thesis, we generate a new dataset and tested it through four Machine Learning (ML) algorithms: Adaptive Boosting (meta. Adaboost (j48)), Random Forest, J48 and MultiLayer Perceptron (MLP). Additionally, we separated the classification process into two cases: Non-Voice Over Internet Protocol (Non-VOIP) and Voice Over Internet Protocol (VOIP) applications, the second one is called Multiclasses, which contains five applications (classes), namely: PayPal, YouTube, Google talk (Gtalk), Yahoo Messenger and Skype.
We choose these applications from the Transport Layer (TL). The generated dataset was compiled by means of a different process that included: packet capturing, features extraction and classification processes, we using also four statistical features. The dataset used here contains real data from a live network using an experimental tested from experimental environment within a campus environment. In the both cases: Non-VOIP and VOIP case and Multiclasses classification case, the meta. Adaboost (j48) classifier achieved the highest accuracy level among other classifiers, of 98.6605% and 98.3007% respectively. The J48 classifier achieved the minimum time for building the training model in the two cases of classification. Also, the MLP took the maximum time between other classifiers for build the training model in both cases.