Machine-Learning Based QoE Prediction For Dash Video Streaming

Tan, Jun Yuan (2021) Machine-Learning Based QoE Prediction For Dash Video Streaming. Final Year Project, UTAR.

Preview

PDF Download (2324Kb) | Preview

Abstract

Quality of experience (QoE) is an essential metric for video service platforms such as Youtube and Netflix to monitor the service perceived by their end-users. Driven by the popularity of MPEG-Dynamic Adaptive HTTP Streaming (DASH) format among service providers, a plethora of QoE prediction models have been proposed for MPEG-DASH video streaming. However, conventional models are established based on machine learning techniques, which are unable to extract high-level features from low-level raw inputs via a hierarchical learning process. The capabilities of deep learning have paved the new way for more powerful QoE prediction models. The aim of this project is to propose a deep-learning-based QoE prediction method. The starting point of the project is a state-of-the-art framework called DeepQoE, which encompasses three phases: feature pre-processing, representation learning and QoE predicting phase. The framework is further improved by integrating ensemble learning in the prediction phase. Extensive experiments are conducted to evaluate the performance of the proposed QoE prediction model as compared to conventional algorithms. By using a publicly available LIVE-NFLX-II dataset, the newly trained model outperforms not only conventional methods but also the DeepQoE by 0.226% and 0.06% in terms of Spearman Rank Order Correlation Coefficient (SROCC) and Pearson Linear Correlation Coefficient (LCC), respectively.

Actions (login required)