01555656

Component

The prior information of travel time assists travelers in choosing appropriate routes and departure times. To achieve this objective, several travel time prediction models have been developed but identification of important predictors and consideration of correlation among them have not received much attention. The authors propose a long-distance freeway travel time prediction and feature selection framework which also accounts for the correlation among covariates. The framework integrates principal component analysis to account for correlation among predictors, neural networks for prediction and a sequential zeroing of weights algorithm for feature selection. As the part of this framework, the authors also propose a straightforward method to retrace the original variables from principal components. To validate this framework, the authors chose a 36.1 km long segment of Taiwan’s National Freeway No. 1 and collected data for forty-three predictors, including rainfall, travel time from the electronic toll collection system, and spot speed and heavy vehicle volume from vehicle detectors. The neural network model predicts travel time on this freeway segment with a mean absolute percentage error (MAPE) of 6.04% using all covariates. The proposed feature selection method further identifies speed and flow of heavy vehicles as the important predictors, whereas rainfall variables as the noisy predictors. The model developed using all covariates excluding rainfall variables is able to predict travel time with a MAPE of 6.09%. The negligible increase (0.05%) in the prediction MAPE after the removal of rainfall variables demonstrates the effectiveness of the proposed feature selection method. These findings facilitate considerable reduction in monetary and logistical expense through equipment saving during future data collection.

This paper was sponsored by TRB committee AHB15 Intelligent Transportation Systems.

01550057

15-1390

English

16p

2015

Transportation Research Board 94th Annual Meeting

Digital/other

Figures; References; Tables

Algorithms; Forecasting; Methodology; Neural networks; Travel time

Taiwan

Data and Information Technology; Highways; I72: Traffic and Transport Planning

Transportation Research Board Annual Meeting 2015 Paper #15-1390

TRIS, TRB, ATRI

Dec 30 2014 12:32PM