Abstract

Worldwide, it is estimated that every year 1.2 million road users lose their lives on road crashes. This situation has a huge impact in terms of health and economical development and costs to governments. Pedestrians and cyclists, often called vulnerable road users (VRUs), are more likely to be injured in road crashes as they are unprotected and more exposed to risk in the presence of motor vehicles. In 2015, 21% of the fatalities in European Union’s roads were pedestrians, while 8% were cyclists. The respective percentages in Portugal were 23% for pedestrians and 5% cyclists. The objective of the paper is to developed a comprehensive study between motor vehicle and pedestrians or cyclists collisions occurrences integrating spatio-temporal data analysis with crash prediction models. Spatial analysis is the first step to identify the patterns between blackspots in three cities from Portugal (Aveiro, Porto and Lisbon). Blackspots were identified using ArcGIS and the Kernel Density Estimation function taking into account the level of injury severity. The second step is a temporal analysis that involves a temporal distribution of the crashes or injured people. Lastly, a crash prediction model was developed for each city that calculates the likelihood of VRUs to be involved in a crash taking into account injury severity. The findings from the study highlighted target variables and specificities at a local level that may influence the number and severity of crashes between motor vehicle and VRUs. The identification of blackspots revealed that most injuries occurs surrounding high attraction places. Temporal analysis results showed a major percentage of crashes occurring during afternoon peak-hours (4-7p.m.). Comparison between the cities showed that Porto presents the worst scenario in terms of number of VRUs injuries per ten thousand inhabitants or kilometers squared. The developed Multinomial Logistic Regression models revealed that VRU gender and age, as well as weather conditions, are statistically significant for the prediction models. The spatio-temporal analysis of crashes taking into account the level of injury severity has the goal of establishing patterns between the different cities, which is important to understand risk factors. The outcomes of the study are valuable not only for VRUs awareness, but also for traffic planners and decision-makers.

Abstract

Vulnerable road users (VRUs) represent a large portion of fatalities and injuries occurring on European Union roads. It is therefore important to address the safety of VRUs, particularly in urban areas, by identifying which factors may affect the injury severity level that can be used to develop countermeasures. This paper aims to identify the risk factors that affect the severity of a VRU injured when involved in a motor vehicle crash. For that purpose, a comparative evaluation of two machine learning classifiers&mdash decision tree and logistic regression&mdash considering three different resampling techniques (under-, over- and synthetic oversampling) is presented, comparing both imbalanced and balanced datasets. Crash data records were analyzed involving VRUs from three different cities in Portugal and six years (2012&ndash 2017). The main conclusion that can be drawn from this study is that oversampling techniques improve the ability of the classifiers to identify risk factors. On the one hand, this analysis revealed that road markings, road conditions and luminosity affect the injury severity of a pedestrian. On the other hand, age group and temporal variables (month, weekday and time period) showed to be relevant to predict the severity of a cyclist injury when involved in a crash. Document type: Article