Fuzzy modeling of freeway accident duration with rainfall and traffic flow interactions

• Accidents duration modeling.
• Augmentation of traffic and environmental variables for accidents׳ duration modeling.
• Fuzzy sets theory for modeling complex traffic data.
• Exploratory analysis for alternative Fuzzy Rule-Based Systems calibration and implementation.
• Comparative analysis among statistical and methods of artificial intelligence.

Accident duration modeling has been considered as a difficult problem due to the variety of information (accident characteristics, traffic and weather information, geometry of the accident location and so on) that should be taken into account to improve predictions and explain the phenomenon. We introduce Fuzzy Rule-Based Systems to model freeway accident duration and cope with the uncertainties and complexities hindering in accident monitoring systems. The models are also compared to classical hazard-based regression models, as well as Multi-Layer Perceptrons. Results show that a Fuzzy Rule-Based System may predict accident duration with fair accuracy using limited information on traffic and weather conditions. Introducing the entire amount of information on accidents to the Fuzzy Rule-Based System leads to reduced modeling accuracy, probably due to the difficulties in converging to a solution. Nevertheless, the Fuzzy Rule-Based System with limited information may predict more accurately than classical hazard based duration models and with comparable accuracy to a Multi-Layer Perceptron, which is presented with information on accident characteristics, traffic and weather conditions, as well as the geometry of the accident location.