Quality of experience prediction in mobility scenarios based on recurrent neural networks.

This paper evaluates the use of machine learning prediction models for assessing the quality of experience perceived by mobile network subscribers under mobility conditions. Input data are taken from a single measuring point available at commercial subscribers’ smartphones. A sequence data classifier based on a type of recurrent neural network, the long shortterm memory network, has been implemented and compared with two simpler machine learning models. A set of nine relevant network parameters, mainly related to the signal level and the signal quality of the serving and the neighbour cells, has been considered as inputs for the prediction model. The prediction model has been applied to estimate the downlink TCP performance on a set of non-intrusive drive tests. The overall performance achieved on the test set is around 80%, a really good result considering the size of the dataset.