Use of machine learning for rate adaptation in MPEG-DASH for quality of experience improvement

Abstract

Dynamic adaptive video streaming over HTTP (DASH) has been developed as one of the most suitable technologies for the transmission of live and on-demand audio and video content over any IP network. In this work, we propose a machine learning-based method for selecting the optimal target quality, in terms of bitrate, for video streaming through an MPEG-DASH server. The proposed method takes into consideration both the bandwidth availability and the client’s buffer state, as well as the bitrate of each video segment, in order to choose the best available quality/bitrate. The primary purpose of using machine learning for the adaptation is to let clients know/learn about the environment in a supervised manner. By doing this, the efficiency of the rate adaptation can be improved, thus leading to better requests for video representations. Run-time complexity would be minimized, thus improving QoE. The experimental evaluation of the proposed approach showed that the optimal target quality could be predicted with an accuracy of 79%, demonstrating its potential.