A gradient boosting procedure in combination with hierarchical reconciliation is proposed in this study for short-term forecasting of traffic flow. Particular attention is paid to three main characteristics of traffic flow: the temporal and spatial patterns, interactions between the temporal and spatial patterns, and the dynamics of traffic flow at different spatial aggregation levels. The performance of the proposed forecasting framework is examined by comparing it with three frequently used methods (i.e., SARIMA, Kalman filter model and random forest) in the literature, and using three distinctive datasets. Overall, the gradient boosting based approach offers a highly flexible and automated way to learn useful information in large datasets, which is particularly advantageous for forecasting traffic flow in a complex road network at longer forecasting horizons.
Short-Term Traffic Flow Forecasting: A Component-Wise Gradient Boosting Approach With Hierarchical Reconciliation
IEEE Transactions on Intelligent Transportation Systems ; 21 , 12 ; 5060-5072
2020-12-01
2194106 byte
Article (Journal)
Electronic Resource
English
Short-term Freeway Traffic Flow Forecasting with ARIMAX Modeling
DataCite | 2010
|Special issue on short-term traffic flow forecasting
Elsevier | 2014
|Special issue on short-term traffic flow forecasting
Online Contents | 2014
|Reliable Short-Term Traffic Flow Forecasting for Urban Roads
Transportation Research Record | 2013
|