Graph Convolution Network for Urban Mobile Traffic Prediction.

Cellular traffic prediction has always been a great challenge in the communication field. Efficiently and correctly predicting the future mobile traffic can improve communication resource’s scheduling. However, changes in mobile traffic are affected by nearby areas. Therefore, prediction methods should be more sensitive to spatial features. Recently, deep learning methods have shown excellent ability in extracting spatial-temporal features. In this work, mobile traffic prediction approach based on the graph convolution networks is propose. We evaluate our model with the Telecom Italia dataset and compare with other traditional models. Experiments show that our model can significantly improve the prediction accuracy.