Ridge penalization-based generalized linear model (GzLM) for predicting risky-driving index

Abstract

Road traffic crashes remain one of the major causes of preventable death and injury worldwide. Human behavior is considered one of the main factors leading to such tragic losses. In this paper, we analyze the responses of an online survey questionnaire and identify the variables that are most likely to be correlated with individual driving behavior of drivers. Weights are allocated to nine risky-driving behaviors considered in the survey based on self-reported frequency of the driving behaviors the participants were involved in at the time of a recent traffic crash. Initially, weighted individual self-rated risky-driving behaviors are used to estimate the risky-driving index (RDI) for individual drivers. RDI is defined as a quantitative measure of a driver's risky-driving propensities based on basic profile and driving history. Finally, a standardized model for predicting a driver's RDI is proposed using Ridge penalization-based generalized linear regression with a standard error of estimate equal to 0.713. According to the model, female drivers have lower RDI compared to male drivers. Also, younger drivers have higher RDI than older drivers. Lastly, hours driven per day have more positive impact on RDI than the number of accidents or the driving experience of a driver. IEOM Society International.