Road Accident Prediction Models and Influence of Traffic Flow, Road Length, Road Class, and Vehicle Class on Accidents

The objective of this paper is to investigate statistical models for road accident occurrence in Great Britain and to identify the potential risk of being involved in accident for a vehicle on the road. We compared the results of statistical models for road accidents developed by generalized linear model and generalized estimation equation techniques with Poisson and negative binomial distributions. The accident occurrence data were extracted from the official records of STATS 19 data from 1999 to 2002. The variables of traffic flow, road length, road class, vehicle class and interaction between traffic flow, road class and vehicle class were used. Durbin Watson test showed the presence of serial correlation in the residuals which affects the significance level of the variables. The Generalized estimation equation with negative binomial regression was preferred because it can accommodate both the over-dispersion and serial correlation that are present in the data. From the results it is found that increase in the traffic flow and road length are associated with more accidents. Car is found to be involved in more accidents than all other vehicle classes reflecting their prevalence. Motorcycle is at more risk per kilometer of travel on motorways, rural minor and urban minor roads than all other vehicle classes whereas pedal cycle is at more risk on rural A and urban A roads. It is also concluded that risk on Urban A roads is higher for all vehicles except goods vehicles in comparison to all other roads. Keywords: Generalized linear model, Generalized estimation equation, Negative binomial regression, Poisson regression, STATS 19 data.