Abstract
With the rapid development of the social economy and accelerating urbanization, the total number of motor vehicles continues to grow at a high rate. Roads in large- and medium-sized cities are becoming increasingly congested, which leads to frequent traffic accidents. To enhance road traffic safety and reduce the traffic accident rate, effectively identifying accident black spots is of great importance. In this study, the data from traffic accidents on the Lianfeng Middle Road, Yinzhou District, Ningbo City were selected for the analytical dataset, and eight impact factors (holiday, day of week, time, rush hour traffic, accident location type, accident type, weather, responsibility and black spot) were set. The improved K-means clustering algorithm was proposed to solve the shortcomings of the traditional algorithm, which is susceptible to outliers and initial clustering centres. Through this algorithm, the traffic accidents in the dataset were divided into two categories: black spots and non-black spots. Then, using the updated dataset, we employed a Bayesian network to construct a black spot identification model, and applied other widely used algorithms (the ID3 decision tree, logistic regression and support vector machine) for comparison. The values of the ROC area, TP rate, FP rate, precision, recall, F-measure and accuracy reached 0.618, 0.668, 0.580, 0.650, 0.668, 0.590 and 0.668, respectively, which showed that the Bayesian network was the best model to effectively identify road accident black spots. Moreover, a bivariate correlation model was applied to verify the correlation between the impact factors and black spots. The results indicated that the accident location type, accident type, time, and responsibility had significant correlations with black spots, which had a value of sig<; 0.05. The conclusions could provide reference evidence for the identification and prevention of traffic accident black spots to significantly contribute to traffic safety.
Highlights
The global status reports on road safety released by the World Health Organization in December 2018 emphasized that the number of road traffic deaths per year has reached 1.35 million [1]
This is consistent with the findings of Yang et al [21], who found that the number of motor vehicles is an important factor in urban road traffic accidents
The Bayesian network (BN) was compared with the ID3, logistic and support vector machine (SVM) algorithms
Summary
The global status reports on road safety released by the World Health Organization in December 2018 emphasized that the number of road traffic deaths per year has reached 1.35 million [1]. In recent years, machine learning algorithms have been widely used in road accident prediction They can effectively classify datasets and establish a link between the factors and the severity of the traffic events. Mbakwe et al [11] combined Delphi technology with BNs to predict road traffic accidents in developing countries Other classification algorithms, such as the ID3 decision tree (ID3), logistic regression (logistic) and support vector machine (SVM), have been employed to classify different types of datasets in previous studies. K-means clustering has been applied to determine the road traffic accident black spots, and BNs have been mostly used to classify and predict traffic accidents. An improved K-means clustering algorithm combined with BNs, ID3, logistic and SVM methods are selected to establish the accident black spot identification models.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
