Investigating the impacts of driver’s risky driving behavior on traffic crash risk detection model

Abstract

This research investigates the impacts of the driver’s risky driving behavior on traffic crash risk detection by constructing four models, including models with and without the use of risky driving behavior data before and during crash. The major findings are as follows. First, the accuracy of the model was improved by adding risky driving behavior variables to the traffic crash risk detection model. In the optimal model, the accuracy, recall, precision, false alarm rate, and the missing report rate are 93.0%, 91.2%, 77.5%, 6.6%, and 8.8% respectively. Second, the impact of traffic flow variables on crash risk in the model was modified by introducing risky driving behavior variables. Among them, the impact of traffic volume on crash risk increases, while the relationship between the average speed, congestion index, and crash risk decreases. Third, the effects of changes in risky driving behavior and traffic flow on the risk of traffic crashes are different at the time slices and space segments, and this effect relationship is characterized by nonlinearity. The results demonstrate the influence of temporal and spatial characteristics of risky driving behaviors on crash risk, which supports detection of traffic crash risks in time and develops crash prevention measures.

Keywords:

• Crash risk detection
• traffic flow
• risky driving behavior
• CatBoost
• SHAP

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Here, $\text{congestion index}=\frac{\mathit{\text{free flow speed}}}{\mathit{\text{average road speed}}},$ e.g., the free-flow state occurs when the congestion index∈[0,1.5), the slow-flow state occurs when the congestion index ∈[1.5,2), the congestion state occurs when the congestion index CI∈[2,4), and the severe congestion state occurs when the congestion index ∈[4,+∞).

2 The calculation for the frequency of events of driving behavior is the driving behavior times of the users in the length l of road i over a period of time, e.g., $\text{frequency of events of driving behavior}=\frac{\mathit{\text{sum}}(\mathit{\text{driving behavior}})}{l_i*\mathit{\text{time}}*\mathit{\text{sum}}(\mathit{\text{users}})}.$

Additional information

Funding

This work was supported by the National Natural Science Foundation of China (No. 52072012), China Postdoctoral Science Foundation (No. 2021M690272), and Research on Key Technologies of Optimization and Safety Assurance Improvement of Eight-lane Freeway Guide System Based on Active Safety (No. 40038001202135).