Short-Term Traffic Flow Prediction Based on Ensemble Machine Learning Strategies

Abstract

In order to solve the problem of traffic congestion, many city governments have begun to develop intelligent transportation systems. As a research hotspot in the field of intelligent transportation, the short-term traffic flow prediction is of great significance to traffic diversion and route planning. In the recent big data era, the machine learning (ML) algorithms have been applied to mining deep information in the data. However, the performance of a single ML model is usually not good. The ensemble learning can improve accuracy in most cases. In this paper, we propose a new short-term traffic flow prediction model. This paper regards the traffic prediction problem as a regression prediction problem, rather than a time series forecasting problem. The proposed model combines the prediction results of the XGBoost, LightGBM and CatBoost models through the ensemble machine learning strategy. With the experiments on real traffic data in Xi'an city, this paper has obtained the comparison of predictive performance of the models. Compared with single models such as the ARIMA, LSTM, XGBoost, LightGBM and CatBoost models, the results show the proposed model is more accurate and more suitable for short-time traffic status prediction.