Modeling Traffic Crashes on Rural and Suburban Highways Using Ensemble Machine Learning Methods

In this research work, clustering techniques were used to characterize crashes occurring on rural and suburban highways in Jordan, while support vector machines were used to compare the classification performance of the resulting clusters with that obtained without clustering, the data used included 10219 crashes obtained from the Jordanian Traffic Institute for (2014–2018). Support vector machines with four kernel functions were used to measure the classification performance using different clustered and non-clustered data. RBF kernel showed a prominent performance in classifying injury severity levels as indicated by the results obtained by the performance measures used (accuracy, precision, recall, and F1-score). It was found that classification performance as measured by support vector machines of the clustered sets was superior to that obtained using the non-clustered set. All clustered models were remarkably able to predict injury severity levels using the RBF kernel with F1-score ranging from 72% to 81%.