Zero Inflated Negative Binomial for Modeling Heavy Vehicle Crash Rate on Indian Rural Highway

Poisson regression and negative binomial regression have been widely used to model the road crashes and to predict crash frequency. Zero inflated models have been shown to be a powerful tool to predict crash frequency when crash data are characterized by preponderance of zero. This paper presents the research work aiming to correlate the road traffic crash rate with road geometry and traffic characteristics for crashes involving heavy vehicles on national highway number 6(NH-6), one of the busy rural roads in central India. Stochastic regression models were developed using crash data collected during 2005-09 over a stretch of 100 km of road length. Zero Inflated Negative Binomial (ZINB) regression method has been used to model the occurrence of road traffic crashes. The Akaike Information Criterion (AIC) has been used to measure the relative goodness of fit. The independent variables selected in this study were shoulder width (SW), lane width (LW), access density (AD), spot speed (SS) and annual average daily traffic (AADT).The model demonstrates that access density, lane width and shoulder width are important parameters affecting the traffic safety of the selected highway.