Spatio-Temporal Recurrent Neural Networks Modeling for Number of Users Prediction on Wireless Traffic Networks

Wireless network traffic modeling is very important for planning, managing, and optimizing computer networks. The dynamic, chaotic, and non-linear nature of network traffic makes an accurate network traffic prediction model very important. The behavior of network traffic which is time series in nature and the existence of location dependencies as well as the linkages between features causes complexity modeling. Therefore, the Spatio-Temporal correlation approach through the Detrended Partial Cross-Correlation Analysis method for feature extraction and Recurrent Neural Networks method is proposed to forecasting a wireless traffic model. The case study for this modeling is to forecast the number of users at three base-stations based on the predictor packets and Bytes variables taking into account the Space-Time effect. The result of the proposed method to forecast the number of users at the three base-stations is that it has a forecasting accuracy of 81%, much higher than the Generalized Spatio-Temporal Auto Regressive of 6.02%.