A hybrid approach for urban expressway traffic incident duration prediction with Cox regression and random survival forests models

Abstract

Traffic incidents such as crashes have significant impacts on urban expressway operation. The roadside service and operational efficiency of urban expressways could be improved based on a well-developed incident duration prediction model. In this study, a hybrid approach that combines Cox regression and random survival forests algorithm is developed to establish incident duration analysis model. The study is conducted based on traffic incident data from Shanghai urban expressways. For each traffic incident, information about the road geometry, traffic operation, and weather conditions was collected for experiments, where 80% of sample is used for training and the rest 20% for validation. In the hybrid model, a Cox regression model is predeveloped to investigate and identify the significant contributing factors of incident duration. Then, these identified significant factors are used as inputs for the random survival forests model. Finally, the statistical measurements including mean absolute error (MAE) and normalized mean square error (NMS) are used to measure the model performance and compare with other models. The analysis results show that incident type, location, affected lane numbers and other attributes have significant impacts on incident duration, and the hybrid approach model provides better prediction accuracy over traditional traffic incident duration prediction methods.