Research Area:  Machine Learning

Abstract:

Machine learning models used in traffic classification make the assumption that the training data and test data have independent identical distributions. However, this assumption might be violated in practical traffic classification due to changes of traffic features. The models trained by existing data will be ineffective in classifying new traffic. A transfer learning model without making the above assumption is proposed in the present study. The maximum entropy model (Maxent) was adopted as the base classifier in the transfer learning model. To examine the efficacy of the proposed method, the traffic dataset collected at the University of Cambridge was used in the condition that the training and test dataset were not identical. Experimental results showed that good classification performance was obtained based on the transfer learning model.

Keywords:  
Network Traffic Classification
Transfer Learning
Machine Learning
Deep Learning

Author(s) Name:  Guanglu Sun,Lili Liang,Teng Chen,Feng Xiaoand Fei Lang

Publisher name:  ELSEVIER

DOI:  10.1016/j.compeleceng.2018.03.005

Volume Information:  Volume 69, July 2018, Pages 920-927

Paper Link:   https://www.sciencedirect.com/science/article/abs/pii/S004579061732829X