Mobile Traffic Prediction Based on AR-GARCH-LightGBM Hybrid Model

Accurate prediction of mobile network traffic is the basis for public network planning, mobile base station management and service quality enhancement. Traditional predictive modeling methods, such as hidden Markov model (HMM) and support vector machine (SVM), mainly focus on the linear time characteristics of traffic data, lacking the capability to address the situation of dense complex networks. In this paper, we propose a hybrid model containing autoregressive moving average (ARIMA) linear prediction and Autoregressive conditional heteroskedasticity mode (GARCH) Residual test, and introduce Light Gradient Boosting Machine (LightGBM) to conduct the network traffic prediction. We decompose the data set into linear and residual parts using DWT. ARIMA model is adopted to deal with the linear component, and the residual parts are predicted by LightGBM. In this process, GARCH is used to test and optimize the prediction of the linear part of ARIMA to improve the prediction accuracy of ARIMA for the linear part. Finally, wavelet reconstruction is used to combine the predicted results. The simulation results show that the prediction accuracy of the hybrid model is improved by 7% compared to the LSTM model, and the time is shortened by 36%.