Traffic analysis on encrypted traffic over wireless channels – Traffic classification based on partial knowledge

PDF

Authors
  1. Song, R.
  2. Willink, T.
Corporate Authors
Defence Research and Development Canada, Ottawa Research Centre, Ottawa ON (CAN)
Abstract
This Scientific Report investigates the limitations of traditional supervised traffic classification techniques on classification performance such as accuracy, precision, and recall when there is only limited knowledge available about the traffic, especially an adversary’s encrypted traffic. To improve the classification performance under the above scenario, a new modified naïve Bayes kernel (MNBK) classifier is proposed based on optimal weight-based (OWB) kernel bandwidth selection. The proposed OWB kernel bandwidth selection algorithm can make a more accurate learning model for traffic classification compared with the traditional classifiers. By generating several possible major traffic types in tactical edge networks, we demonstrate that the proposed MNBK classifier not only improves the classification performance on the existing classes significantly, but also detects unknown traffic with very high accuracy, precision, and recall compared with the traditional classifiers. In addition, a learning classification model is proposed based on MNBK, that processes received ongoing real time traffic and updates the classification table periodically. Generally speaking, with more and more accurate information retrieved from received real time traffic, the proposed real time classification model should improve the classification performance over time compared with the traditional classifiers that do not consider the ongoing received traffic. This has been demonstrated with our c

Il y a un résumé en français ici.

Keywords
Traffic Analysis;Traffic Classification;Machine Learning;Gaussian Kernel Density;Bayes Theorem;Naïve Bayes Classifier;Naïve Bayes Kernel Estimation
Report Number
DRDC-RDDC-2018-R196 — Scientific Report
Date of publication
01 Nov 2018
Number of Pages
46
DSTKIM No
CA047642
CANDIS No
808157
Format(s):
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: