Investigation of Young Pedestrian Crashes in School Districts of New Jersey Using Machine Learning Models

Approximately half of all traffic-related casualties across the globe involve vulnerable road users such as pedestrians, bikers, and bicyclists. The National Highway Traffic Safety Administration (NHTSA) reported that in the United States, crashes involving school buses resulted in a total of 111 fatalities in 2019. Specifically, almost one-sixth of the pedestrians involved in motor vehicle crashes in New Jersey over the last five years (2016-2020) were less than 18 years of age. Although various initiatives have been implemented to boost the safety of school children, crashes involving school children are still worrisome, leading to an emerging national issue. Finding unknown patterns in a complex, multivariate crash data set is often difficult for supervised algorithms, as they rely on a limited number of predetermined premises. Unlike supervised algorithms, unsupervised algorithms can find the relevant inherent trends between the variables and crash severity. Thus, to figure out the causes of these young pedestrian crashes, this study obtained and investigated five years of crash data sets (2014-2019) in New Jersey school districts using a multilayer artificial neural network. Shapley Additive Explanation of SHAP values of the top contributing factors is further performed to assess the impact of those factors on the crash severity. The analysis will help interpret the potential causes of the crashes, suggest countermeasures, and create awareness in combating pedestrian crashes in school districts.