Traffic Prediction Based on Ensemble Machine Learning Strategies with Bagging and LightGBM

With the development of mobile networks, one of the main challenges is performing accurate prediction in order to maximize resource usage, saving energy and improving quality of service (QoS). In the recent big data era, Machine learning (ML) algorithms have been exploited to mine the profound information hidden in the data that are suitable for describing the instability of network traffic, but the performance of a single ML model is often not very good. Ensemble learning can further increase accuracy on a variety of ML tasks. Therefore, in this paper, we apply random forest (RF) and LightGBM to mobile network traffic prediction by using RF to filter redundant features and using LightGBM to train prediction model. Furthermore, we propose a new traffic prediction model based on ensemble framework of bagging and LightGBM. The proposed model is evaluated with a real-life traffic dataset. The experiment results show that the proposed model effectively improves the prediction performance compared to single LightGBM given the same number of decision trees and some other popular algorithms, ARIMA, multi-layer perceptron (MLP) and Linear Regression (LR).