A machine learning model for qoe prediction of dynamic video streaming

Abstract

Quality of Experience (QoE) is one of the important factors in deciding the efficiency of a network provider. This work discusses about the methodology in which machine learning can be incorporated to predict QoE of a dynamic streaming video or a webcam capture. The prediction of QoE is done using the combination of simpler neural network and Deep learning technique. The neural network trained with the H.264 encoded bit-streams is the Non-Linear Autoregressive Exogenous (NARX) model. Along with this unsupervised phase of training, Deep learning of Restricted Boltzmann Machine (RBM) is done keeping in mind the time-series changes of a streaming video. The NARX model acts as the training model that is capable of capturing the events from the database. Its feature of feedback network is extracted to make the prediction more apt. RBM deep learning extracts information from the trained neural network to predict PSNR, RMSE and SSIM which is in par with the ground truth. The combination of these algorithms gives prediction near to ground truth. The dataset used is encoded in the H.264 codec format. The presented work has very effective application in meeting the customer gratification by the network provisionary.