Content area
Full text
njury and death rates from vehicle crashes are high for most parts of the world, with major influences including driver, vehicle, roadway, and environmental characteristics.1 For road characteristics, recent research reported that fatality index-ratio is higher for curved and at-grade segments (rolling terrain roads) than non-rolling terrain roads.2 Furthermore, Federal Highway Administration (FHWA) reports, that more than a quarter of all crashes occur at horizontal curves where vehicles depart from the roadway and hit trees, utility poles, rocks, or other permanent features.3 Constrained topography and complex road geometries are among major issues in designing and constructing roads that meet appropriate standards. Substandard cross-section elements and dangerous roadside environments plus human driver's errors in these areas present a risky road traffic situation for drivers.
Previous research shows that most crashes on horizontal curves occur when drivers are speeding or distracted.4 Also, the magnitudes of radius of adjacent horizontal curves and tangent lengths between curve were associated with departure crashes.6 Hummer et al., identified these crashes as associated with presence of fixed objects, such as trees and poles; influence of alcohol on drivers; and lighting and roadway surface conditions.7 Other studies show a higher number of crashes occur at curved segments than tangent segments, and more crashes occur at two-lane than multilane roads.4 Severity of injuries caused by these crashes varies with radius and length of the curves, where curve radius is negatively correlated with occurrence of crashes.10 Authors identified 2,500 feet (762 meters) as threshold curve radius below which rate of crash occurrence increases.10 Rusli et al., reported that steep gradients and sharp curves along rolling terrain roads induce different driving behavior than curved roads in flatter areas.2
In light of this review, there remain significant gaps where research is warranted. Most studies in the literature either look at curved or grade alignments; few studies have analyzed safety factors of segments with a combination of the two. Moreover, most of these studies used traditional statistical models which are limited by overfitting and poor performance on non-linear relationship of variables. In addressing these limitations, current study investigates effects of curve and grade combination segments, with other traffic characteristics, in predicting drivers' injury severity. This is achieved by a comparative assessment of robust predictive machine learning models (Random...