QoE-aware ML models based on network parameters for video streaming over 5G O-RAN architecture

This work proposes a machine learning (ML) based solution for enabling mobile network operators (MNOs) to estimate the quality of experience (QoE) provided during a video session just from the network's side data. The proposed ML model inserted into the 5G Open Radio Access Network (O-RAN) enables addressing multiple end devices (EDs) with a QoE-aware resource allocation without leveraging on the Service Provider (SPs) for the QoE estimation. We assume that the SP and the MNO cooperate during the training process, labeling the network-based collected dataset. The resulting ML model estimates the average QoE during the overall session time, which is influenced by the EDs' mobility behavior, the user channel quality variations, and available network resources. The proposal introduces a unified tool for addressing fixed and mobile EDs requesting videos with resolutions up to 4K and frame rates up to 60 fps. Multiple supervised ML regression models were trained and tested, where Gradient Boosting (GB) achieved the highest QoE estimation performance (R2 = 0.986, RMSE = 0.091).