Estimation of Traffic Incident Duration: A Comparative Study of Decision Tree Models

Unexpected events such as crashes, disabled vehicles, flat tires and spilled loads cause traffic congestion or extend the duration of the traffic congestion on the roadways. It is possible to reduce the effects of such incidents by implementing intelligent transportation systems solutions that require the estimation of the incident duration to identify well-fitted strategies. This paper presents a methodology to establish incident duration estimation models by utilizing decision tree models of CHAID, CART, C4.5 and LMT. For this study, the data contained traffic incidents that occurred on the Istanbul Trans European Motorway were obtained and separated into three groups according to duration by utilizing some studies about classification of traffic incidents. By using classified data, decision tree models of CHAID, CART, C4.5 and LMT were established and validated to estimate the incident duration. According to the results, although the models used different variables, the decision tree models of CHAID, CART and C4.5 have nearly the same prediction accuracy which is approximately 74%. On the other hand, the prediction accuracy of decision tree model of LMT is 75.4% which is somewhat better than the others. However, C4.5 model required less number of parameters than the others, while its accuracy is the same with others.