Crash prediction based on traffic platoon characteristics using floating car trajectory data and the machine learning...

Area of Science:

• Traffic Safety Engineering
• Transportation Systems Analysis
• Data Science in Transportation

Background:

• Accurate crash prediction on urban expressways is crucial for safety improvements.
• Existing methods struggle with real-time data collection and fail to incorporate traffic platoon characteristics.
• Traffic platoons significantly influence crash dynamics, yet are often overlooked in predictive models.

Purpose of the Study:

• To develop a model for predicting crash propensity using traffic platoon characteristics.
• To evaluate the effectiveness of the floating car method for collecting relevant traffic data.
• To compare the performance of binary logistic models and support vector machines in crash prediction.

Main Methods:

• Collected crash and floating car data from Shanghai's Middle Ring Expressway.
• Implemented a data preparation method including filtering and network matching.
• Applied binary logistic regression and support vector machine (SVM) models for analysis.

Main Results:

• Traffic platoon information from floating cars is effective for expressway crash prediction.
• The SVM model achieved 85% accuracy, significantly outperforming the binary logistic model (60% accuracy).
• The binary logistic model offers better interpretability regarding factors contributing to crashes.

Conclusions:

• Floating car technology and SVMs show promise for real-time crash prediction on expressways.
• While SVMs excel in predictive accuracy, logistic models provide valuable insights into crash contributing factors.
• Integrating platoon characteristics enhances the accuracy of expressway safety analysis.