Developing a QoE Monitoring Approach for Video Service Based on Mobile Terminals

Quality of experience (QoE) is crucial for wireless network operators with respect to both technical evolution and profit promotion. However, operators are still unable to monitor user QoE effectively because traditional network quality indices are not directly related to the user’s audiovisual experiences. Motivated by the fact that mobile devices are the network elements closest to the users, we explore the possibility of estimating user QoE by capitalizing on mobile terminal capabilities. For this purpose, we first collect over eighty thousand data records through a self-developed mobile application operating in the real world conditions with the support of a prominent Chinese wireless network operator (i.e., China Unicom). The collected data consist of 89 objective parameters concerning wireless video services and 4 types of subjective user scores. Then, we perform data preprocessing and feature selection based on Spearman correlation analyses. Finally, we establish two predictive models for QoE estimation based on the classification tree C4.5 algorithm and the gradient boosting decision tree (GBDT) algorithm, respectively. The experimental results demonstrate that the two decision tree-based models outperform other machine learning algorithms. Specifically, the prediction accuracy of the GBDT is approximately 70% for a five-level scale and approaches 90% for a more practical 3-level scale. Therefore, this study strongly demonstrates the feasibility of the user terminal-based approach for QoE prediction.