Determination of Various Risk Elements’ Association Causing Fatal Crashes in Heterogeneous Traffic Conditions

Abstract

Historically, in an accident event, a single parameter on its own is hardly ever responsible for crashes. Two or more attributes combine to create a compound effect, which results in vulnerability and creates or causes the possibility of a road accident. Therefore, a study on composite effects is needed to address the many influencing characteristics of an accident. At the same time, it is also essential to know how features alone or in combination with other features increase the likelihood of an accident. A literature review found that crash analysis studies have considered traffic or non-traffic characteristics as effective factors in fatal or non-fatal crashes in previous studies. No studies have considered all the possible attributes together and analyzed their effect on fatal as well as non-fatal crashes. At the same time, there is a lack of research examining the individual and combined effects of traffic, non-traffic, physical, built-environment factors, and road user demographic factors on fatal/non-fatal crashes. In this context, the primary objective of this study is to examine the prevalence of elements associated with and influencing fatal and non-fatal collisions among all parameters. Therefore, this research presents a comprehensive methodology for identifying critical risk elements or attributes associated with fatal road crashes in Indian road conditions. This study addresses this gap by assessing these factors using police registry data and estimating the crude odds ratio (OR) to identify risks associated with fatal collisions. In this regard, logistic and log-binomial regression models are used to estimate the OR and confidence interval (CI) associated with the risk elements for exploring the interrelationships between a set of risk factors (traffic and built environmental attributes) and fatal crash occurrences; such variables were based on the associated OR and CI estimates. Critical risk elements are identified and subsequently divided into three groups: important risk elements, inconsequential risk elements, and risk-free elements. In the subsequent stage, the data mining algorithms Apriori and FP-Growth were adopted to study the combined effect of two or more risk elements leading to significant numbers of fatal crashes. The use of the FP-Growth data mining algorithm to explore risk factor interrelationships remains one of the major contributions of this study. Further, results derived from logistic regression models and data mining techniques are combined to classify the overall set of attributes into five different classes, namely factors with: 1) extreme dominance, 2) strong dominance, 3) moderate dominance, 4) low dominance, and 5) extremely low dominance risk elements. Such risk element classification approach and identification of key risk variables would aid in formulating targeted risk mitigation measures for improved road safety.