01338069

Component

In this paper, the authors propose a new promising machine learning technique to select important explanatory covariates, as well as to improve crash prediction; the group least absolute shrinkage and selection operator (GLASSO) technique. GLASSO’s main strength lies in its ability to deal with datasets having relatively large number of categorical variables, which is the case in this study. Identifying the significant factors affecting safety of unsignalized intersections was also an essential objective. Two applications of GLASSO were investigated; application for variable screening before fitting the traditional negative binomial (NB) model, as well as before fitting another promising data mining technique (the multivariate adaptive regression splines “MARS”). Extensive data collected at 2475 unsignalized intersections were used. For fitting the NB models, the backward deletion and the random forest techniques were separately used as variables screening, and their prediction performance was compared to that from GLASSO. All the three methods resulted in almost similar predictions. For GLASSO’s second application with MARS, the model fitting relatively outperformed that from the random forest technique with MARS, with similar prediction performance. Due to its outstanding performance with categorical variables, as well as its simplicity, GLASSO is recommended as a promising variable selection technique. Significant predictors affecting total crashes at unsignalized intersections were traffic volume on the major road, the upstream and downstream distances to the nearest signalized intersection, median type on major and minor approaches, and type of land use. Resembling previous studies, the volume of traffic was the most important predictor.

01329018

11-0091

English

20p

2011

Transportation Research Board 90th Annual Meeting

DVD

Figures (1) ; References (41) ; Tables (6)

Algorithms; Data mining; Highway safety; Land use; Mathematical prediction; Signalized intersections; Traffic volume

Nonsignalized intersections

Data and Information Technology; Highways; Safety and Human Factors; I80: Accident Studies

Transportation Research Board Annual Meeting 2011 Paper #11-0091

TRIS, TRB

Feb 17 2011 5:19PM