基于自动机器学习的QoE预测模型研究

Research on QoE prediction model based on automatic machine learning

准确预测流媒体视频的用户体验质量(QoE)是提升其服务的关键所在。流媒体QoE预测模型通常基于视频质量和网络传输质量等客观指标进行评估,然而,QoE的主观性为准确评估带来了极大的挑战。为了更精确地预测用户体验质量,文章首次将自动机器学习用于流媒体视频的QoE预测,提出了基于自动机器学习的QoE预测模型。该模型通过特征分析从视频质量评估指标和网络质量评估指标中选择最优特征作为输入,采用H2O AutoML自动机器学习算法进行QoE建模。为了评估方法的有效性,在公开数据集SQoE-Ⅲ数据库上进行实验,并与基于传统机器学习的XGBoot算法的QoE模型结果进行对比分析。实验结果显示,通过自动选择和调优,基于自动机器学习的QoE预测模型取得了显著的进展。该模型的MAE为5.53699、RMSE为7.35987,有效提升了QoE预测的准确性。该研究为QoE建模提供了新的思路和方法,精确预测了用户对视频流的感知满意度。

Accurately predicting the users’quality of experience(QoE)for streaming video is crucial for enhancing its service.QoE prediction models for streaming media typically rely on objective metrics such as video quality and network transmission quality.However,the subjectivity of users’QoE poses significant challenges for accurate assessment.In order to more precisely predict user experience quality,this paper introduces,for the first time,the application of automated machine learning to QoE prediction for streaming video,proposing an automated machine learning-based QoE prediction model.The model utilizes feature analysis to select optimal features from video quality assessment metrics and network quality assessment metrics as input,employing the H2O AutoML automated machine learning algorithm for QoE modeling.To evaluate the effectiveness of the method,the experiments are conducted on the publicly available SQoE-Ⅲdatabase,comparing the results with a traditional machine learning XGBoost-based QoE model.The experimental results demonstrate the significant progress in QoE prediction by adopting the automated machine learning-based model through automatic feature selection and model tuning.The model’s MAE is 5.53699,and RMSE is 7.35987,effectively improving the accuracy of QoE prediction.Therefore,this study provides new perspectives and methods for QoE modeling,contributing to a deeper understanding and precise prediction of user perceptual satisfaction with streaming video.