Content-Based Video Quality Prediction Using Random Neural Networks for Video Streaming over LTE Networks

Quality of experience (QoE) varies dramatically according to the video content. The most common video quality prediction models are based on the application and network layers. However, little research has been done to predict video quality on the basis of video content over fourth-generation (4G) wireless networks. The aim of this study is to classify video content according to the impact of video content on video quality in streaming H.264 video over long-term evolution (LTE) networks, using 'cluster analysis'. This classification is used to develop novel content-based prediction models using random neural networks (RNNs) for video applications. The content based prediction models are then trained with a gradient descent (GD) training algorithm for the four distinct content types and tested using an untrained dataset. Our content-based prediction models are used to establish the relationship between network, application, and LTE related layers to video quality for all video content types. The proposed video prediction model performed well in terms of the root mean squared error (RMSE) and Pearson correlation coefficient an about 12% increase compared to fuzzy logic and adaptive neural fuzzy models. Our simulation results demonstrated that the proposed scheme provides good predictive accuracy (~ 93%) between the measured and predicted values. This work can potentially help in developing accurate and effective reference-free video prediction models and admission quality of service (QoS) control mechanisms for video streaming over LTE networks.