engleski

YouTube QoE Estimation Based on the Analysis of Encrypted Network Traffic Using Machine Learning

The widespread use of encryption in the delivery of Over-The-Top video streaming services poses challenges for network operators looking to monitor service performance and detect potential customer perceived Quality of Experience (QoE) degradations. While monitoring solutions deployed on client devices provide insight into application layer KPIs (e.g., video quality levels, buffer underruns, stalling duration) which can be further mapped to user QoE, network providers commonly rely primarily on passive traffic monitoring solutions deployed solely within their network to obtain insight into user perceived degradations and their potential causes. In this paper we present a methodology for the estimation of end users' QoE when watching YouTube videos which is based only on statistical properties of encrypted network traffic. We have developed a system called YouQ which includes tools for monitoring and analysis of application-layer KPIs and corresponding traffic traces, and the subsequent use of this data for the development of machine learning models for QoE estimation based on traffic features. To test this approach, we have collected a dataset of 1060 YouTube video traces using 39 different bandwidth scenarios. All video traces are annotated with application-layer KPIs and classified into one of three QoE classes. The dataset was used to test various machine learning algorithms, and results showed that up to 84% QoE classification accuracy could be achieved using only features extracted from encrypted traffic.

Quality of Experience ; YouTube ; Machine learning ; encrypted network traffic

nije evidentirano