Fatal crashes and rare events logistic regression: an exploratory empirical study

Area of Science:

• Traffic Safety Analytics
• Statistical Modeling
• Transportation Engineering

Background:

• Fatal road accidents are infrequent events, challenging traditional statistical models.
• Accurate prediction of fatal crashes is crucial for effective traffic safety interventions.

Purpose of the Study:

• To evaluate the effectiveness of a rare events logistic model (RELM) for predicting fatal road accidents.
• To compare the predictive accuracy of RELM against the classic logit model (LM).

Main Methods:

• Both logit model (LM) and rare events logistic model (RELM) were applied to analyze fatal crash data.
• Crash-injury datasets from Hillsborough County, Florida, were utilized for empirical evaluation.
• Model performance was assessed using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) metrics.

Main Results:

• The rare events logistic model (RELM) demonstrated superior accuracy in predicting fatal crashes compared to the classic logit model (LM).
• Empirical analysis, supported by higher AUC values, indicated a clear advantage for RELM in predictive performance.

Conclusions:

• The rare events logistic model (RELM) is a more proficient tool for predicting fatal crashes than the classic logit model (LM).
• RELM is recommended for advanced traffic safety analytics requiring precise estimation of rare events.