This is an interim version of our Electronic Legal Deposit Catalogue-eJournals and eBooks while we continue to recover from a cyber-attack.
Traffic-Condition-Awareness Ensemble Learning for Traffic Flow Prediction⁎This work was partially supported by the National Key R&D Program of China (No. 2018YFB1700202), and the National Natural Science Foundation of China (No.61903363, No.U1811463, and U1909204), and the Early Career Development Award of State Key Laboratory for Management and Control of Complex Systems (No.20190205). Issue 5 (2020)
Record Type:
Journal Article
Title:
Traffic-Condition-Awareness Ensemble Learning for Traffic Flow Prediction⁎This work was partially supported by the National Key R&D Program of China (No. 2018YFB1700202), and the National Natural Science Foundation of China (No.61903363, No.U1811463, and U1909204), and the Early Career Development Award of State Key Laboratory for Management and Control of Complex Systems (No.20190205). Issue 5 (2020)
Main Title:
Traffic-Condition-Awareness Ensemble Learning for Traffic Flow Prediction⁎This work was partially supported by the National Key R&D Program of China (No. 2018YFB1700202), and the National Natural Science Foundation of China (No.61903363, No.U1811463, and U1909204), and the Early Career Development Award of State Key Laboratory for Management and Control of Complex Systems (No.20190205).
Abstract: Traffic prediction is an elemental function of Intelligent Transportation Systems (ITS), and accurate and timely prediction is significant for both proactive traffic control and providing traveler information. In this paper, we focus on investigating ensemble leaning that benefits from different base models, and propose a traffic-condition-awareness ensemble approach. We apply graph convolution on the network of traffic detectors to capture the spatial patterns embedded in traffic flow. Then, the extracted features are used to formulate a weight matrix to ensemble the predictions of base models according to their performances under a certain condition. We performed a series of experiments on a real dataset to compare the proposed methods with several competitive models, including ensemble methods: Weight Regression model and Gradient Boosting Regression Tree model, and single model approach: Support Vector Regression (SVR), Long Short-term Memory (LSTM) model and Historical Average Model model. Experimental results demonstrate that our method can significantly improve the performances of traffic flow prediction.