Current freeway traffic flow prediction techniques pay attention to time series prediction or introduce the upstream adjacent road segments in the short-term prediction model. In this paper, all of the road segments on the freeway are considered as candidates of the independent variables fed into the prediction model. A spatio-temporal multivariate adaptive regression splines (MARS) approach is proposed for the road network analysis and to predict the short-term traffic volume at the observation stations on the freeway. The actual traffic data are collected from a series of observation stations along a freeway in Portland every 15 minutes. In the first phase, the macroscopic dependency relationships of the stations on the freeway are investigated via MARS method. Subsequently the stations most related to the object station are selected and fed into the MARS prediction model to generate the short-term volume. The experiments are carried out on the actual traffic data and the results indicate that the proposed spatio-temporal MARS model can generate superior prediction accuracy in contrast with the historical data based MARS model, the parametric ARIMA, and the nonparametric PPR methods.
A spatio-temporal multivariate adaptive regression splines approach for short-term freeway traffic volume prediction
2013-10-01
875114 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Short-Term Traffic Prediction considering Spatial-Temporal Characteristics of Freeway Flow
DOAJ | 2021
|Using multivariate adaptive regression splines (MARS) in pavement roughness prediction
Online Contents | 2003
|Short-Term Freeway Traffic Flow Prediction: Bayesian Combined Neural Network Approach
Online Contents | 2006
|Stochastic Approach for Short-Term Freeway Traffic Prediction During Peak Periods
Online Contents | 2013
|