Ignoring Encrypted Protocols: Cross-layer Prediction of Video Streaming QoE Metrics

Nowadays, the widely used end-to-end encryptions have prevented the Deep Packet Inspect (DPI) method applied in Quality of Experience (QoE) prediction and optimization. Thus, for the Internet Service Providers (ISPs), it has become more difficult to monitor and improve user QoE when users are watching videos or using other video application services. To overcome the above problem, a Cross-layer Predicting framework of Video Streaming QoE (CPVS) is proposed to monitor and predict video indicators that affect user QoE, such as startup delay, stalling rate and stalling events. The most prominent advantage of the CPVS is that it can predict the video QoE indicators without the prior information of transmission protocols, whether it is encrypted or not. Furthermore, the general framework, CPVS, shows high predicting accuracy both in a time granularity as small as one second and in video-session prediction, even can perform well with only 6 features. For the session-based prediction, the best weighted f1 score achieves 85.71% about the startup delay prediction, and achieves 84.64% about the stalling rate classification. For the real-time prediction, the best weighted f1 score can achieve 79.25%. Besides, by exploring the 4G network features, we find out the 6 most critical ones that not only make the CPVS retain high accuracy but also save 35.02% running time to improve the real-time prediction performance. © 2022, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.