Spatial variable selection methods for network-wide short-term traffic prediction
Md. Mahmud Hasan, Jiwon Kim, Carlo Giacomo Prato · Transport Research Forum · 2017
This paper proposes statistical approaches to identifying spatial relationships among road links in an urban road network to select predictors for a short term traffic prediction model for a given road link more systematically. For this purpose, two methods i.e. Granger causality test and elastic net regularization techniques have been adopted in this study. Urban road network in Brisbane, Australia is selected as a study site for a case study. One-year traffic flow and speed data from the selected road network have been used for the analysis. The study evaluates the performance of the proposed variable selection methods in terms of the prediction accuracy of short term traffic prediction models constructed based on the variables identified by the methods. This study uses time lagged multiple linear regression method as a short term traffic prediction model. For a given target link, the relevant predictors obtained by Granger causality and elastic net are used separately to build the respective traffic prediction models. It is observed that Granger causality based traffic prediction model provides better prediction accuracy than elastic net based traffic prediction model.