چكيده به لاتين
Firstly, the present study is aimed to evaluate influencing factors on accident frequency using ANOVA test in Boroujerd-Khoramabad four-lane rural road in order to road safety risk index based on K-means clustering and accident frequency. The present study is also aimed to apply principle components analysis (PCA) as a static prioritization tool and Gaussian probability-density model as a dynamic prioritization tool for identifying the main and sub-main factors contributed to accident frequency. Furthermore, several statistical models such as multinomial logistic regression (MNL), and ordered probit (OP) models and particle swarm optimization (PSO) as a feature selection are applied for optimizing the number of variables and improving the prediction performance of crash severity in DT techniques including gradient boosting (GB) and random forest (RF) models. In the present study, the influencing factors on road accidents including human, roadway, monitoring, traffic, development factors in Boroujerd-Khoramabad four-lane rural road are examined. using K-means clustering and chi-square are applied in order to categorize drivers based on the accident risk. Furthermore, two prioritization models including PCA model as a static prioritization tool and Gaussian probability-density model as a dynamic prioritization tool for prioritizing influencing factors contributed to road accident frequency are applied. Moreover, the data records of crashes are examined regarding the crash severity such as fatality, injury, and property damage only (PDO) in rural roads. Then, factors contributed to crash occurrence are investigated regarding multinomial logistic regression (MNL), and ordered probit (OP) models and in DT techniques based on particle swarm optimization (PSO). The findings indicated that operating speed and the differences between posted speed limits and operating speed are the pivotal influencing factors on accident frequency rate. The results showed that operating speed and the difference between posted speed limits and operating speed are the most influencing factors on accident frequency. Moreover, the results of K-means clustering analysis showed that six clusters with accident risk were identified as highly, relatively highly, moderately, relatively lowly, lowly risky, and not risky (safe) clusters. Regarding the K-means clustering analysis, the accident frequency is increased by decreasing the difference between the posted speed limits and the operating speed from the safe cluster. In addition, the safety risk index model based on the Gaussian model indicated that the average reducing factor of accident frequency reached 0.99 by increasing 1 km/hr in the difference between the posted speed limits and the operating speed among low risky and safe clusters, while it was equal to 1.17 in risky and unsafe clusters. Therefore, the increasing factor of accident frequency in in risky and unsafe clusters are 1.18 times than the ones in low risky and safe clusters. Additionally, the maximum and minimum values of safety risk index were observed in the sixth and the third clusters, respectively. the results of PCA model indicated that posted speed limits, operating speed, the difference between posed speed limits and operating speed, segment length, and traffic flow, respectively. However, findings from the dynamic prioritization model as Gaussian probability-density model indicated that the most critical factor was known as segment length. Furthermore, traffic flow, posted speed limits, difference between posted speed limits and operating speed, and operating speed were ranked as other critical factors which contribute to accident frequency, respectively. A comparison between the static and dynamic prioritization models revealed that the dynamic prioritization model could be more reliable than the static method due to considering the probability and density of accident frequency.The results of using PSO into DT techniques represented that the most variables having a role in crash occurrence are cause of the crash, crash type, and weather conditions, speed limits, and the number of vehicles involved in the crash, respectively. In addition, the PSO-GB model with eight essential variables demonstrated higher prediction performance in comparison with other variables. Regarding crash severity prediction, it was found that PSO-GB improved the prediction performance by 96.37, 95.02, and 92.21% for PDO, injury, and fatality, respectively, leading to an estimation of the overall accuracy of nearly 94.01%. Based on the sensitivity analysis, the PSO-GB model is the best predictive model while OP is the weakest one.
