In the present study, the author has extracted the data from 1255 cases of accidents investigation hardcopy reports at Ajdabiya traffic police center. Due to the lack of computerization in the traffic police center, the data were archived as a hard copy format, and the accident investigation reports were handwritten, without uniform form, and no data coding. For that, the data collection was conducted manually, and then Microsoft Excel was used to construct and to code an electronic database from 2001 to 2010. The database was constructed based on temporal and spatial factors of the accident and contained detailed information about the vehicle, driver, passenger, type of accident, cause of the accident, and accident injury severity. For modeling purposes, accident severity was classified into two levels of the KABCO scale: minor injury and severe or fatal injury [8]. Other variables, including the age of the victim, spatial distribution, temporal distribution, type of accident, cause of the accident, and vehicle type, were taken into consideration in this study. Only 565 of RTAs occurred on El-Brega Coastal Freeway were considered in this study.

Descriptive analysis of the data was conducted using the two-way frequency table to examine the association between the independent variables and accident severity. Data were analyzed using STATA 15 software. However, high frequency does not mean they are statistically significant. For example, vehicle Type “passenger car” has a high rate of accidents occurring but has no statistical significance in the final model.

The logistic regression model is a statistical method that is frequently used by road safety researchers to analyze data when investigating the association between various factors and crash severity. In this study, injury severity in crashes was defined as a binary variable. A binary logistic regression model was developed to predict accident injury severity in El-Brega Coastal Freeway by using the binary model in the analysis.

BLM can be formed as follows:

Pi=Prob. yi= yi/Xi is the response probability to be modeled and yi is the first ordered level of y

The odd ratio for dichotomous explanatory variable x, which takes a value of 1 or 0 (1 indicates that the event will absolutely occur and 0 indicates that the event will absolutely not occur), can be expressed as the ratio of the predicted number of times when an event will occur (x = 1) to the predicted number of times it will not occur (x = 0). The flowing formula can illustrate this condition:

Where

= Odd ratio

π1/1-π1 = Odds that the event will occur when x = 1

π0/1-π0 = Odds that the event will not occur x = 0

As shown in Table 3, only Twenty-three variables were first considered for this study. At first, the Univariable analysis has been done to test the associations between the predictors and the outcome, a level of <0.10 is chosen as the level of significance to identify variables known to be important. The variables with an effect on crash injury severity at a level of significantly higher than 90% (i.e., α ≥ 0.90) were included in the model development process. Second, collinearity was checked to identify the associated collinear pairs for retaining only one of the variables. Third, the stepwise backward elimination method was followed to select a set of predictors, only statistically significant variables at the p= 0.05 level should stay in the model. Finally, Goodness-of-fit (GOF) statistics were used to evaluate Model adequacy, and the Hosmer-Lemeshow test result shows that the model fits well with p-value = 0.2435. Table 4 presents the final model with significant predictors. STATA 15 software was utilized in the analysis.

Although the weekly distribution of road accidents in Fig. (4a) shows that more accidents occurred on weekdays. However, the model result shows that most crashes happened on the weekend are more severe than those that happened in weekdays with OR = 1.72. This finding is similar to that of Yu & Abdel-Aty and Shrestha & Shrestha [9, 10]. For instance, Shrestha & Shrestha [10] attributed that most drivers traveling on weekends are less familiar with the roads. Regarding daytime and night, descriptive analysis in Fig. (4b) shows a slightly difference between accidents frequency and severity. Moreover, day time is not significant in the final model.

The graph in Fig. (4c) shows that the horizontal curvature contributes to increasing accident frequency; also, horizontal curves represent 43% of Accident black spots, as shown in (Table 2). In final mode, injury severity at the horizontal curve rises by 65% with OR= 1.65, as well as same is the case in the most previous researches that have confirmed that accidents occur more frequently and more severely on a horizontal curvature [11].

Crash investigation reports listed 15 types of first harmful events. In the present study collisions were sorted into six categories. The first five categories of collisions were a rollover, head-on, accidents involving camels, sideswipe-opposite, and rear-end. The sixth category, which comprised all other types of collisions, had a low percentage of occurrences. Several other studies have used this accident classification Kim et al. and Tavris et al. [12, 13]. Referring to the Descriptive analysis results Fig. (4d), we can see clearly that rollover collision is the most frequent. Nevertheless, the final model result shows that rollover collision is 47% less severe than other accidents types with (OR = 0.53). In the descriptive analysis, head-on collisions accounted for only 14% of total accidents. In spite of this, modeling result (Table 4) shows that head-on collisions influencing accidents severity in El-Brega freeway by an odds ratio of (OR = 3.28), indicate the probability of severe injuries in head-on collisions which is 228% higher compared to other types of accidents. This finding is consistent with those of Deng et al.; Şimşekoğlua et al.; and Hosseinpour et al [14-16]. Deng et al. [14] found that fatal head-on crashes are much higher on undivided rural two-lane highways than on other types of roadways.

A rear-end collision is less severe than other accidents types, as shown in the Table 2. The odds ratio of rear-end accidents is 0.27. Our result in line with the previous finding; for example, Kim et al. [12] reported similar findings that rear-end collisions rarely lead to a severe injury. Camel accidents and Sideswipe-O do not have any statistical significance.

Police investigations reported the causes of accidents in 42 categories. The percentages for some categories were very low and were therefore combined into four groups, namely, driver behavior, environmental-related causes, road-related causes, and vehicle-related causes. Several essential causes of accidents were listed separately, such as high speed, tire burst, and animals, as illustrated in Fig. (4e). Even so High-speed caused only 23% of accidents. However, it has statistical significance in the final model result, as shown in (Table 4), indicating high speed which leads to severe injuries probability by 116% with OR = 2.16. This result quite agreed with the literature which concluded that speed had been known to have a high likelihood of causing the most severe accidents [17, 18]. For example, Da Silva et al. [18] found that high speed increased the risks of occurrence and the severity of accidents. High-speed driving has been known to have a high probability of causing the most severe accidents [17]. Descriptive analysis shows that animals cause a small portion of accidents. And the odds ratio of involved in vehicle accidents is OR= 0.45, which means that accidents caused by animals are less severe compared to other causes. If we know that, the vast majority of animals in this study are camels; the finding is inconsistent with the previous results. For instance, Al Shimemeri and Arabi [19] stated that big animals (i.e., camel) involved in traffic collisions usually result in severe injuries to passengers. In the present study, the vast majority of animals are camels. We argue that this contradiction may be related to the difference in drivers behavior from country to others; in such situations, drivers may have familiarity with camels grazing areas. Although descriptive statistics in Fig. (4e and f) show that driver behavior made up the highest percentage of accidents caused, and about 60% of these accidents were severe. However, this factor has no statistical significance in term of accident severity. Additionally, environmental-related causes, road-related causes, and vehicle-related causes, and tire burst did not have statistical significance in term of accident severity.

Fig. (4g) shows that driver of the age group 25-40 tends to have the highest accident frequency, however the severity ratio is almost the same as in all age groups (50%). In addition, in the model result, driver’s age seems to be a significant factor, nevertheless, had a minor effect on accident severity, as each change in the aged unit (year) is accompanied by 2% (OR= 1.2) increase in the severity of the crash. This finding is somewhat consistent with the most common conclusion in the previous researches, the increasing of driver age increase the severity of the injury [20, 21].