While there have been many studies analyzing crash severity, few studies have accounted for temporal transferability and compared different crash severity models. This study attempts to investigate the contributing factors for the injury severity of pedestrian-involved crashes for different time of day (daytime vs. nighttime). To achieve this, a large sample of crash records from 2010 to 2014 (5 years) from the state of California was collected from the Highway Safety Information System (HSIS) crash dataset. A multinomial logit (MNL) modeling approach is applied to determine the statistically significant injury severity contributing factors by time of day. In addition, a parameter transferability test is conducted to determine if daytime crashes and nighttime crashes need to be considered separately for safety analyses. A Support Vector Machine (SVM) model is also employed for prediction performance comparison purposes.
The empirical results reveal that factors, including pedestrian action, type of vehicle involved, roadway type, weather condition, and accident type, are among the top three impactful variables for fatal and severe injury levels for daytime and nighttime crashes. In addition, the results show that SVM models provide superior results compared to MNL model.
Mokhtarimousavi, S. (2019). A Time of Day Analysis of Pedestrian-Involved Crashes in California: Investigation of Injury Severity, a Logistic Regression and Machine Learning Approach Using HSIS Data. ITE Journal, 89(10), pp 25-33.