Road Accident Case Modeling in Ghana: A Perspective of Negative Binomial–Based Generalized Linear Models

Abstract

The gravity of casualties due to road traffic accidents poses a threat to national development, and this requires pragmatic and actionable road traffic policy and regulation restructuring. This restructuring in policy and regulations can be best achieved through appropriate statistical modeling framework in the space of large road traffic accident data in which hidden data issues are inevitable. In this regard, this paper investigates the perspective of negative binomial generalized linear models (NBGLMs) with and without LASSO regularization in modeling road traffic accident casualties in Ghana, in which predictor space data issues are inevitable. Implementation using real road traffic accident casualty data compiled by Building and Road Research Institute in Ghana from 2017 to 2021 illustrates the potential of the LASSO negative binomial generalized linear model (LNBGLM) over its unregularized counterpart NBGLM in terms of both fitting and predictive performances. Model assessment was quantified in statistics such as deviance, AIC, BIC, and graphical predictive performance analysis. In particular, results show that LNBGLM is more suitable than NBGLM in estimating road accident casualties in Ghana with type of weather, vehicle involved in an accident, location, and age of victims as major determinants. The practical implications of the results become apparently clear and realistic, examining the state of the identified predictors on our roads currently.