Traffic Forecasting With Uncertainty: A Case for Conformalized Quantile Regression
G. Petelin, J. Rožanec, G. Papa
European Congress on Computational Methods in Applied Sciences and Engineering ECCOMAS 2024
Lisbon, Portugal, 3-7 June, 2024
Accurate and reliable traffic flow forecasting is of primary importance for traffic planning and management. While there is a growing interest in real-time traffic forecasting models, accurate predictions remain a challenge due to the dynamic nature of traffic systems and the multiple factors that affect the traffic flow. Point forecasts do not provide insights regarding uncertainties associated with forecasts. Furthermore, many traffic flow models fail to produce prediction intervals that accurately capture the uncertainty of the forecasts. Therefore, we investigate the use of quantile regression models[1] for traffic flow forecasting and highlight their tendency to generate prediction intervals that are too narrow and poorly calibrated[2]. To address this issue, we propose using conformal predictions, which allow us to achieve well-calibrated prediction intervals leading to more accurate, reliable, and therefore trustworthy predictions. Additionally, we show that using quantized conformal regression to calibrate machine learning models offers much more accurate predictions that do not deviate significantly even for longer forecasting horizons. This advanced approach enhances the robustness of the forecasts by effectively addressing the uncertainties inherent in the traffic system, thus providing a more reliable tool for traffic management and planning. The authors acknowledge the financial support of the Slovenian Research and Innovation Agency under research core funding No. P2-0098. This work is also part of a project that has received funding from the European Union’s Horizon Europe research and innovation program under Grant Agreement No. 101077049 (CONDUCTOR).
BIBTEX copied to Clipboard