Identifying the Severity of Road Accident Impact on Traffic Flow by Ensemble Model

Road accidents are one of the prime reasons for traffic congestions and delays. Predicting the severity of traffic congestions caused by road accidents can help to prepare travelers and traffic managers for the delays and re-routings. This study constructs a Stacked Ensembled Classifier model (SEC) by combining three ensemble machine learning algorithms Random Forest (RF), Extreme Gradient Boosting (XGB) and Gradient Boosting Machines (GBM). Severity of the impact of a road accident on the current traffic flow is predicted based on time, weather, distance, location, road conditions, traffic signal and other points of interests. These three ensemble models are chosen because they achieved higher accuracies than other baseline models. The results indicate that our stacked classifier outperforms individual performance of the ensemble algorithms by up to 3.2%. Additionally, the average daily traffic count on each road segment, whose data is obtained from Virginia department of transportation, is introduced here as a new input feature. Inclusion of this feature improves the prediction accuracy up to 4.6%.