A conflict-based approach for real-time road safety analysis: Comparative evaluation with crash-based models

Area of Science:

• Transportation Engineering
• Traffic Safety
• Machine Learning Applications

Background:

• Traditional real-time crash prediction models (RTCPMs) rely on historical crash data, limiting their application in data-scarce environments.
• Accurate, high-resolution spatio-temporal crash data is often unavailable or unreliable for real-time road safety analysis.
• Surrogate safety measures offer an alternative for assessing road safety without relying solely on crash data.

Purpose of the Study:

• To introduce and evaluate a novel real-time conflict prediction model (RTConfPM) for road safety analysis.
• To assess the efficacy of RTConfPM trained with surrogate safety measures compared to traditional RTCPMs.
• To predict the risk of unsafe situations, specifically rear-end crashes, in advance using real-time traffic data.

Main Methods:

• Developed an RTConfPM using time-to-collision (TTC) values from radar sensors to define unsafe situations.
• Utilized traffic conditions as input variables for the RTConfPM.
• Employed Pearson's correlation test and random forest for variable selection, SMOTE for class balancing, SVM for classification, and Monte Carlo cross-validation for robustness.

Main Results:

• The conflict-based approach (RTConfPM) significantly outperformed the traditional crash-based approach (RTCPM).
• RTConfPM achieved over 93% accuracy, recall, and specificity in predicting unsafe situations within a 5-minute interval.
• The model demonstrated superior performance in terms of accuracy, recall, specificity, and AUC compared to RTCPM.

Conclusions:

• The proposed RTConfPM, trained with surrogate safety measures like TTC, is a promising approach for real-time road safety analysis.
• This method effectively addresses limitations posed by the unavailability or unreliability of historical crash data.
• RTConfPM offers a robust and accurate tool for proactive identification of potential road safety hazards.