Page 26 - tmp_Neat

binary probit model to investigate the effect of under-reported collision data and found that

sequential logit performs better than the standard ordered probit model in addressing the
parameter bias estimation due to under-reported data.

2.2.2.10 Artificial Neural Networks (ANNs)

A neural network model allows for potentially non-linear relationships between the injury

severity and independent variables (Salvolainen et al. 2011). Chimba and Sando (2009)
developed an artificial neural network (ANN) model to examine crash injury severity and

found the ANN is able to predict crash injury severity with higher accuracy than traditional
ordered probit models. The optimisation of the number of hidden neurons will increase the

prediction accuracy of ANN models.

2.3 Heavy Vehicle Crash and Injury Severity

Many studies have investigated factors contributing to the frequency and severity of crashes
involving heavy vehicles (Yasmin et al., 2010; Mooren et al., 2014). The main variables

considered in previous research have been roadway characteristics (e.g. number of lanes)

(Islam, 2015; Islam and Hernandez, 2016), traffic conditions (e.g. traffic volume) (Duncan et
al., 1998; Lee and Li, 2014), temporal characteristics (e.g. time of day) (Islam et al., 2014;

Lee and Li, 2014; Marquis & Wang, 2015; Pahukula, 2015), environmental factors (e.g. light
conditions) (Pahukula et al., 2015; Islam and Hernadez, 2016), vehicle characteristics (e.g.

vehicle type) (Lemp et al., 2011; Lee and Li, 2014), collision characteristics (e.g. sideswipe
crashes) (Pahukula et al., 2015; Islam & Hernadez, 2016) and occupant characteristics (e.g.

driver age) ( Zhu and Srinivasan, 2011b).

Miaou (1994) investigated the effect of road alignment on the frequency of heavy vehicle

collisions on a rural interstate freeway in the American state of Utah. The factors considered

were speed limit, annual average daily traffic (AADT) per lane, horizontal curvature, vertical
grade, shoulder width and percentage of heavy vehicles. Dong et al. (2015) investigated the

effect of geometric design on crashes occurrence involving heavy vehicles on highways in
Tennessee, USA. They showed that the risk of crashes between cars and trucks was higher in

commercial areas, while the risk of crashes between heavy vehicles was higher in industrial
zones.

15