Procjena performansi usluge YouTube na osnovu analize kriptiranog prometa uporabom tehnike strojnog učenja

master's thesis

• Title (english): YouTube performance estimation based on the analysis of encrypted network traffic using machine learning
• Author: Irena Oršolić
• Mentor: Lea Skorin-Kapov (mentor)
• Committee member: Lea Skorin-Kapov (predsjednik povjerenstva)
• Committee member: Maja Matijašević (član povjerenstva)
• Committee member: Mirko Sužnjević (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2016-07-07, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

Ovaj rad nastao je u okviru projekta koji se bavi procjenom iskustvene kvalitete usluge YouTube na osnovu analize kriptiranog mrežnog prometa. Tokom projekta, razvijen je sustav YouQ, u svrhu prikupljanja podataka na klijentskom uređaju (Android pametnom telefonu) i u mreži. Sustav čini i YouQ poslužitelj za obradu i prikaz podataka. Fokus ovoga rada je na drugoj fazi projekta: korištenju algoritama strojnog učenja nad podacima prikupljenim sustavom YouQ kako bi se procjenila klasa iskustvene kvalitete video strujanja.

Analizirani skup podataka sastoji se od 343 primjera opisana s 54 značajke mrežnog prometa i klasom iskustvene kvalitete ("high", "medium", "low"). Korištenjem "wrapper" metoda u alatu WEKA odabrani su podskupovi značajki za četiri algoritma strojnog učenja: Naive Bayes, SMO, J48, i Random Forest. Izgrađeni su klasifikacijski modeli korištenjem algoritma OneR i četiri prethodno navedena algoritma. Točnost klasifikacije je procijenjena i raspravljena. Rezultati pokazuju da je najtočniji model onaj stvoren koristeći algoritam Random Forest, s točnošću 74%. Najviše pogrešaka u procjeni dogodilo se između klasa "low" i "medium", u eksperimentima s promjenama mrežne propusnosti.

Osim opisa metodologije manipulacije značajkama, izgradnje i evaluacije modela, ovaj rad pruža teoretske informacije o algoritmima strojnog učenja, kratki uvod u korištenje WEKA alata i statističku analizu podataka prikupljenih na klijentskom uređaju i u mreži. Također, kako bi skup podataka korišten za učenje bio što jasniji, opisano je laboratorijsko okruženje i metodologija prikupljanja podataka.

Abstract (english)

This thesis was completed in the scope of a project focused on YouTube QoE prediction based on encrypted network traffic. During this project, a system called YouQ was developed for the purpose of collecting data both on a client device (Android smartphone) and in the network. The system also includes the YouQ server for processing and visualizing data. This thesis focuses on the second phase of the project: using machine learning algorithms on data collected by the YouQ system to predict QoE classes of each YouTube video streaming session.

The dataset that was analysed consists of 343 instances described with 54 network traffic features and a QoE class label ("high", "medium", "low"). The dataset was subset using wrapper methods in the WEKA machine learning tool for four machine learning algorithms: Naive Bayes, SMO, J48, and Random Forest. Classification models were created using OneR and the four listed algorithms. Classification accuracy was assessed and discussed. The results show that the most accurate was the model created using the Random Forest algorithm, with 74% accuracy. Most of the errors happened between classes "low" and "medium" and in the experiments with bandwidth fluctuations.

Besides presenting the methodology of feature manipulation, model creation, and evaluation, this thesis provides theoretical information about algorithms, a short guide to using WEKA, and statistic analysis of both application-level and network-level data. Further, to better understand the dataset that was used as an input to machine learning algorithms, the experiment setup and data collection methodology has been described.

• Keywords: klasifikacija mrežnog prometa; strojno učenje; iskustvena kvaliteta; YouTube
• Keywords (english): network traffic classification; machine learning; Quality of Experience; YouTube
• Language: english
• URN:NBN: urn:nbn:hr:168:847619
• Study programme: Title: Information and Communication Technology; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije (magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2018-02-06 11:00:34