Predicting crash injury severity in road freight flows with association rules algorithms

Abstract

The purpose of this study is to evaluate the use of the Apriori association rule mining algorithm to classify and predict the severity of the 718,565 accidents involving freight transport vehicles in Mexico, which occurred between 2009 and 2018. The accidents were classified into those in which there was only material damage or injured people {Severity=0} and in those in which people died {Severity=1}. 115 association rules were obtained, 79 corresponding to non-fatal accidents, and 36 to fatal ones. The main factors associated with the severity of the accident belong to male subjects, involved in accidents that occur on weekends and in suburban areas, and where the probability of the accident being fatal is 1.69 times greater. Thus, the results of using the association rules to relate demographic and circumstantial characteristics of the accident with the severity of the injuries show an accuracy of just over 65%. Therefore, despite the limitations that may occur due to the omission of relevant variables, and the fact that the results show little precision, the feasibility of using machine learning techniques and, specifically, the association rules as promising tools to help analyze accidents and help launch road safety interventions more effectively is manifested.