Heterogeneous ensemble learning for enhanced crash forecasts - A frequentist and machine learning based stacking...

This study aims to increase the prediction accuracy of crash frequency on roadway segments that can forecast future safety on roadway facilities. A variety of statistical and machine learning (ML) methods are used to model crash frequency with ML methods generally having a higher prediction accuracy. Recently, heterogeneous ensemble methods (HEM), including "stacking," have emerged as more accurate and robust intelligent techniques providing more reliable and accurate predictions.

This study applies "Stacking" to model crash frequency on five-lane undivided (5 T) segments of urban and suburban arterials. The prediction performance of "Stacking" is compared with parametric statistical models (Poisson and negative binomial) and three state-of-the-art ML techniques (Decision tree, random forest, and gradient boosting), each of which is termed as the base-learner. By employing an optimal weight scheme to combine individual base-learners through stacking, the problem of biased predictions in individual base-learners due to differences in specifications and prediction accuracies is avoided. Data including crash, traffic, and roadway inventory were collected and integrated from 2013 to 2017. The data are split into training (2013-2015), validation (2016), and testing (2017) datasets. After training five individual base-learners using training data, prediction outcomes are obtained for the five base-learners using validation data that are then used to train a meta-learner.

Results of statistical models reveal that crashes increase with the density (number per mile) of commercial driveways whereas decrease with average offset distance to fixed objects. Individual ML methods show similar results - in terms of variable importance. A comparison of out-of-sample predictions of various models or methods confirms the superiority of "Stacking" over the alternative methods considered.

From a practical standpoint, "stacking" can enhance prediction accuracy (compared to only one base-learner with a particular specification). When applied systemically, stacking can help identify more appropriate countermeasures.