Razvoj modela temeljenih na strojnom učenju za procjenu performansi usluge YouTube na osnovu značajki kriptiranog prometa

undergraduate thesis

• Title (english): The Development of Machine-Learning Based Models for Estimating YouTube Performance Based on Features Derived from Encrypted Traffic
• Author: Josip Pečarić
• Mentor: Lea Skorin-Kapov (mentor)
• Committee member: Lea Skorin-Kapov (predsjednik povjerenstva)
• Committee member: Irena Oršolić (član povjerenstva)
• Committee member: Ivan Bartolec (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-12, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Promet video strujanja danas čini glavninu ukupnog prometa na Internetu. Kako bi se izbjegao problem zastoja u prijenosu video sadržaja, usluge video strujanja počele su koristiti standard prilagodljivog video strujanja. Također, usluge video strujanja kriptiraju svoj promet, među kojima je i usluga YouTube. Zbog kriptiranog prometa u svojoj mreži, pružatelji internetskih usluga nemaju uvid u kvalitetu video strujanja sadržaja kod krajnjih korisnika. Rješenje je izrada modela temeljenih na strojnom učenju koji se treniraju na značajkama prometa snimljenog s mrežne razine te njima pripadnim izračunatim ključnim pokazateljima performansi s aplikacijske razine, koji su snimani na mobilnim uređajima krajnjih korisnika.
U sklopu ovog rada, izrađeni su modeli procjene temeljeni na algoritmima klasifikacije. Modeli su istrenirani nad već snimljenim i prikupljenim podacima. Nad modelima su napravljene metode odabira značajki i podešavanje hiperparametara, s ciljem povećanja točnosti procjene modela. Također je prikazan utjecaj veličine skupa podataka na točnost procjene modela.
Dobiveni rezultati upućuju na zadovoljive točnosti procjene kvalitete izrađenih modela. Pokazano je kako odabirom značajki i podešavanjem hiperparametara povećava se točnost modela. Zaključeno je kako povećanjem veličine skupa podataka može dovesti do boljih rezultata procjene modela.

Abstract (english)

Today, video streaming traffic makes up for most of the total Internet traffic. To avoid stalling occurrences, video streaming services started using the HTTP adaptive streaming paradigm. Also, video streaming services are employing traffic encryption, including the YouTube service. Because of the encrypted traffic in its network, Internet service providers have no insight into video stream quality for end users. A potential method of solution is to build machine learning models that train on network features and their calculated key performance indicators (KPIs) from the application level that were recorded on end-user mobile devices. Such models can then estimate KPIs solely based on the analysis of encrypted network traffic.
As part of this paper, estimation models are based on classification algorithms. The models have been studied over the already recorded and collected data. Feature selection and hyperparameter tuning were set up to increase the accuracy of model estimations. It is also showed how the size of dataset has an impact on model accuracy.
The obtained results indicate a satisfactory accuracy for estimated values which indicate video quality. It has been shown that feature selection and hyperparameter tuning can increase accuracy of the model. It has been concluded that increasing the size of the dataset can lead to better model estimation results.

• Keywords: YouTube; prilagodljivo strujanje putem protokola HTTP; protokol QUIC; strojno učenje; procjena iskustvene kvalitete; procjena performansi
• Keywords (english): YouTube; HTTP adaptive streaming; QUIC; Machine Learning; Quality of Experience; Performance Evaluation
• Language: croatian
• URN:NBN: urn:nbn:hr:168:681031
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 21:08:55