Traffic incidents adversely affect the safety and mobility of our transportation network. As such, accurate prediction of incident duration is critical in developing strategies and deploying resources to quickly clear incidents and restore traffic to pre-incident conditions. This study introduces a mixture density network (MDN) based on Gamma and Weibull distributions to estimate incident clearance duration. The MDN is known for being highly flexible and can recognize multiple components in the distribution of a target variable. A total of 58,167 incidents from highways in Jacksonville, Florida, gathered from 2014 to 2017, were used as a case study. The comparison between MDN, basic artificial neural network (ANN), and XGBoost revealed that the MDN outperformed the other models by having the lowest mean square error (MSE) and mean absolute error (MAE). The MSE of Gamma MDN was the lowest, estimated at 926 min, compared to 935, 945, and 975 min of the Weibull MDN, ANN, and XGBoost, respectively. Based on the Gamma MDN, the key variables influencing incident clearance duration estimated by the permutation feature importance and shapley additive explanations algorithms include the type of agencies that responded to incidents, the number of agencies involved, incident type, and incident severity. The practical contribution and the application of this study in diverse areas have been discussed. It is expected that the findings will help to improve incident clearance strategies. Specifically, the developed model could be utilized to develop incident management strategies that will proactively address the safety and mobility impacts of traffic incidents on roadways.

    The findings from this study are useful to researchers, traffic operators, and incident responders to develop effective and proactive traffic incident management strategies. Researchers can develop an artificial intelligence algorithm with MDN to predict clearance duration based on real-time traffic and weather data. Traffic operators and incident management teams can use these results to strategically allocate appropriate incident response resources along freeway segments while understanding the impact of surrounding conditions when clearing incidents. Besides, the model could be used to proactively inform drivers upstream of the traffic incident, depending on its length of clearance duration. This information would enable drivers to take proactive measures, including detouring in case of incidents with overly long clearance duration.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Prediction of Traffic Incident Clearance Duration Using Neural Network for Multimodal Data Distribution


    Weitere Titelangaben:

    J. Transp. Eng., Part A: Systems


    Beteiligte:


    Erscheinungsdatum :

    2024-09-01




    Medientyp :

    Aufsatz (Zeitschrift)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch




    Big data-driven highway traffic incident duration prediction method

    CHEN JIAONA / LI DAOFENG / TAO WEIJUN et al. | Europäisches Patentamt | 2022

    Freier Zugriff

    Freeway incident duration prediction using Bayesian network

    Yang, Hongtai / Shen, Luou / Xiang, Yunchi et al. | IEEE | 2017


    Traffic Control Concepts for Incident Clearance

    I. Birenbaum / C. Creel / S. G. Wegmann | NTIS | 2009



    Competing risks mixture model for traffic incident duration prediction

    Li, Ruimin / Pereira, Francisco C. / Ben-Akiva, Moshe E. | Elsevier | 2014