| چکیده انگلیسی مقاله |
According to the World Health Organization report, Iran ranks 113th out of 175 countries with an estimated rate of 20.5 deaths per 100,000 population due to crashes. Additionally, road crash injuries are considered one of the top five leading causes of mortality in the country. Due to the efficiency of intelligent methods, statistical models have attracted a lot of attention in safety studies and crash analysis. Investigations show that few studies have been done regarding the specific effects of time periods of the day on the crash severity. Moreover, in most conducted researches for modeling crash severity, statistical models have been utilized, and determining the relationships between severity and influential factors with these models is challenging due to the complex relationships between variables that vary in different time intervals throughout the day. In recent studies, road lighting conditions have been considered as an important parameter in the frequency and severity of driving crashes. The lighting conditions, especially on rural roads, depend on various factors such as drivers' visibility to perceive road geometry, traffic, and roadside objects. The variables considered for analyzing crash severity include human factors, weather conditions, primary causes of crashes, intersection control, land use, lighting conditions, location of crash, time of crash (day, month, year, and hour), type and condition of road surface, road markings, road repairs, collision type, and road geometry. Therefore, the purpose of this research, while considering the interpretability of the Ordered Probit algorithm, is to investigate the effect of the conditions of different times of the day and night on the crash severity with six years data of Lushan-Qazvin rural freeway crahses. The Ordered Probit algorithm was implemented and evaluated separately in four time periods to predict the effects of various geometric, environmental and traffic factors on the severity of damage caused by crashes and the likelihood ratio test was used to justify the models. The results obtained from the research not only contribute to predicting preventive measures for crashes but also aid in reducing their severity and analyzing the relationships between various factors. One of the stages in conducting research on crash injuries involves identifying important factors affecting the severity of injuries. The estimation results indicate that the levels of crash severity in different time intervals are associated with various factors in different ways, and these differences cannot be discovered by estimating only one general model. The present study has shown that by interpreting machine learning algorithms, useful patterns can be extracted from a large volume of crash data and analyze crash severity. The ordinal probit algorithm is one of the powerful tools for modeling crash severity, which can demonstrate the complex relationship between input variables and crash severity by providing statistics such as z-test and marginal effect in the model output. The analysis results indicate that driver distraction during the morning, noon, evening, and night periods, with coefficients of 1.33, 1.21, 1.25, and 0.98 respectively, is the most significant factor contributing to the increased crashes severity on the studied road. Conversely, collisions with fixed objects during the same periods, with coefficients of -0.58, -1.83, -1.24, and -0.75 respectively, have an inverse relationship with the crashes severity. |