Modelling trends in road crash frequency in Qatar State

Abstract

The data-based regression models are widely popular in modelling the relationship between the crash frequencies and contributing factors. However, one common problem usually associated with the classical regression models is the multicollinearity, which leads to biased estimation of the model coefficients. This paper mainly focuses on the consequences of multicollinearity and introduces a multiple objective-based best-subset approach for promoting the accuracy of the road crash model in Qatar State. The prediction performance of the methodology is verified through a comparative study with two of well-known time series models, namely autoregressive moving average (ARMA) and double exponential smoothing (DES). The mean absolute percentage error (MAPE) is used to assess the ability of each model in maintaining minimum prediction errors. The methodology is illustrated by using a data set of road crashes in Qatar State, 2007-2013.