Predicting road accidents: a rare-events modeling approach

Modeling road accident occurrence has gained increasing attention over the years. So far, considerable efforts have been made from researchers and policy makers in order to explain road accidents and improve road safety performance of highways. In reality, road accidents are rare events. In such cases, the binary dependent variable is characterized by dozens to thousands of times fewer events (accidents) than non-events (non-accidents). Instead of using traditional logistic regression methods, this paper considers accidents as rare events and proposes a series of rare-events logit models which are applied in order to model road accident occurrence by utilizing real-time traffic data. This statistical procedure was initially proposed by King and Zeng (2001) when scholars study rare events such as wars, massive economic crises and so on. Rare-events logit models basically estimate the same models as traditional logistic regression, but the estimates as well as the probabilities are corrected for the bias that occurs when the sample is small or the observed events are very rare. Consequently, the basic problem of underestimating the event probabilities is avoided as stated by King and Zeng (2001). To the best of our knowledge, this is the first time that this approach is followed when road accident data are analyzed. Instead of applying a traditional case-control study, the complete dataset of hourly aggregated traffic data such as flow, occupancy, mean time speed and percentage of trucks, were collected from three random loop detectors in the Attica Tollway ("Attiki Odos") located in Greater Athens Area in Greece for the 2008-2011 period. The modeling results showed an adequate statistical fit and reveal a negative relationship between accident occurrence and the natural logarithm of speed in the accident location. This study attempts to contribute to the understanding of accident occurrence in motorways by developing novel models such as the rare-events logit for the first time in safety evaluation of motorways. Language: en

[1]  Wei Wang,et al.  Evaluation of the impacts of traffic states on crash risks on freeways. , 2012, Accident; analysis and prevention.

[2]  Vikash V. Gayah,et al.  Crash Risk Assessment Using Intelligent Transportation Systems Data and Real-Time Intervention Strategies to Improve Safety on Freeways , 2007, J. Intell. Transp. Syst..

[3]  Gary King,et al.  Toward a Common Framework for Statistical Analysis and Development , 2008 .

[4]  Gary King,et al.  Logistic Regression in Rare Events Data , 2001, Political Analysis.

[5]  Mohamed Abdel-Aty,et al.  Bayesian Updating Approach for Real-Time Safety Evaluation with Automatic Vehicle Identification Data , 2012 .

[6]  Mohamed Abdel-Aty,et al.  Identifying crash propensity using specific traffic speed conditions. , 2005, Journal of safety research.

[7]  Mohamed M. Ahmed,et al.  The Viability of Using Automatic Vehicle Identification Data for Real-Time Crash Prediction , 2012, IEEE Transactions on Intelligent Transportation Systems.

[8]  Kara M. Kockelman,et al.  Freeway Speeds and Speed Variations Preceding Crashes, Within and Across Lanes , 2010 .

[9]  Veerle Vanacker,et al.  Logistic regression applied to natural hazards: rare event logistic regression with replications , 2012 .

[10]  Mohamed M. Ahmed,et al.  Assessment of Interaction of Crash Occurrence, Mountainous Freeway Geometry, Real-Time Weather, and Traffic Data , 2012 .

[11]  Gary King,et al.  Explaining Rare Events in International Relations , 2001, International Organization.

[12]  Mohamed Abdel-Aty,et al.  Bayesian random effect models incorporating real-time weather and traffic data to investigate mountainous freeway hazardous factors. , 2013, Accident; analysis and prevention.

[13]  Gary King,et al.  Zelig: Everyone's Statistical Software , 2006 .

[14]  Mohamed M. Ahmed,et al.  Exploring a Bayesian hierarchical approach for developing safety performance functions for a mountainous freeway. , 2011, Accident; analysis and prevention.