A comparative study of models for the incident duration prediction

Abstract

Purpose This study is intended to investigate the reliability of different incident duration prediction models for real time application with a view to contribute to the development of a decision aid tool within the incident management process context where rough incident duration estimates are currently provided by traffic operators or police on the basis of their skill and past experience. Methods Five predictive models, ranging from parametric models, to non-parametric and neural network models, have been considered and compared evaluating their capacity of predicting incident duration. The data set used in this study for developing and testing the prediction models includes 237 incident events and contains information about the incident characteristics, the personnel and equipment involved to clear the incident and the related response times, including the beginning and ending time of the incident. Results Testing results have demonstrated that the proposed models are able to achieve good performance in terms of prediction accuracy especially for incidents with duration less than 90 min. This finding is partly due to the fact that the dataset has a relatively small number of severe incidents. Furthermore a linear combination of predictions from models was applied with negligible gain in accuracy. Conclusions A deeper investigation is suggested for a future work to evaluate potential improvements from the application of other combination methods. Moreover each proposed model is able to reach best performance for incidents within a particular duration range. Thus a preliminary incident classification scheme could be more convenient in order to select the more appropriate prediction model.

Analysed potential of big data and supervised machine learning techniques in effectively forecasting travel times from fused data

“…The k-nearest neighbours (kNN) is a non-parametric method that can be used for both regression and classification tasks [19,20,21,22]. As our goal is to forecast travel time (continuous dependent variable) the focus will be primarily on the kNN regression.…”

Modelling the impact of traffic incidents on travel time reliability

“…Hence, it was concluded that the results of this model were not promising. Valenti et al (2010) compared the results of five predictive models including KNN in order to investigate the reliability of different incident duration prediction models. The results showed that this method was inclined to overestimate the incident cases with relatively short durations and conversely underestimate incident cases with relatively long durations.…”

“…It indicated that the poor performance of the forecasting models was due to the poor quality of the accident data. Valenti et al (2010) developed five predictive models to investigate the reliability of different incident duration prediction models. The results of validation of the DT model indicated that the model was not capable of performing accurately for durations of more than 90 minutes and less than 5 minutes.…”

“…However, the prediction results were better for incident cases with the duration between 20 and 30 minutes. Valenti et al (2010) indicated that 79% of SVM prediction errors were less than 20 minutes. Moreover, the SVM model exhibited the best prediction for medium and mediumlong duration incidents.…”

“…The model related to crash type provided slightly better accuracy than the two other incident types. Although the accuracy of the duration models are not ideal, these are assessed as reasonable predictions based on the MAPE scale as illustrated in Chapter 4, by Kang and Fang (2011), 44% by Hu et al (2011), and between 34% to 44% for different modelling approaches by Valenti et al (2010). (1) percent of incidents within a given difference between actual and predicted…”