Quality of Experience Prediction of HTTP Video Streaming in Mobile Network with Random Forest

As video is witnessing a rapid growth in mobile networks, it is crucial for network service operators to understand if and how Quality of Service (QoS) metrics affect user engagement and how to optimize users' Quality of Experience (QoE). Our aim in this paper is to infer the QoE from the observable QoS metrics using machine learning techniques. For this purpose, Random Forest is applied to predict three objective QoE metrics, i.e., rebuffering frequency, mean bitrate and bitrate switch frequency, with the initial information of each video session. In our simulation, QoE of four different video streamings are analyzed with eight different system loads. Results show that sufficient prediction accuracy can be achieved for all QoE metrics with the attributes we adopted, especially with low and middle system loads. In terms of type of streamings, the prediction of all metrics for static users performs better than mobile users. Feature selection is also implemented under the highest load to examine the effect of different attributes on each QoE metric and the correlation among attributes.