Loading...

Online High-bandwidth Network Application Detection Using Stream Classification

Nazari, Ziaeddin | 2019

1563 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 51903 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Jalili, Rasool
  7. Abstract:
  8. Trac classication in today’s high-bandwidth networks is challenging, resource consuming, and inaccurate due to the high volume, velocity, and variety aracteristics of the network trac. Trac aracterization and Application identication teniques are widely addressed in the current literature. Due to the massive volume and streaming data in recent years, stream algorithms have been considered by many researers in dierent areas. Online application detection is an issue that has been addressed less frequently in literature. In this thesis, we investigate the performance of 10 dierent stream classication algorithms along with traditional classication algorithms. To generate a robust classier for identifying applications over encrypted trac, we proposed DSCA as a DPI-based Stream Classication Algorithm. DSCA utilizes applications detected by DPI as labels for incoming trac ow. Accordingly, the classi cation models are adapted to consider new labeled data. To reduce the classication algorithms overhead without accuracy reduction, DSCA uses a feature selection method, named CfsSubsetEval. DSCA is applied on UNB ISCX VPN-nonVPN dataset and UNB ISCX Tor-nonTor dataset via MOA tool. e simulation results show that the DSCA with Adaptive Random Forest stream classication approa has the best performance over UNB ISCX VPN-nonVPN whi processed the dataset in 8.63 seconds with 96.75%. About UNB ISCX Tor-nonTor dataset, DSCA with Knn with PAW classication approa has the best performance (86.92% accuracy and 12 seconds execution time)
  9. Keywords:
  10. Network Traffic ; Data Stream Classification ; Application Identification ; Incremental Learning ; Traffic Analysis ; Online Detection

 Digital Object List

 Bookmark

...see more