A Comparative Study on the Performance of Evolutionary Fuzzy and Crisp Rule Based Classification Methods in Congestion Prediction

Accurate estimation of the future state of the traffic is an attracting area for researchers in the field of Intelligent Transportation Systems (ITS). This kind of predictions can lead to traffic managers and drivers to act in consequence, reducing the economic and social impact of a possible congestion. Due to the inter-urban traffic information nature, the task of predicting the future state of the traffic requires, in most cases, a non-linear patterns search in the input data. In recent years, a wide variety of models has been used to solve this problem in the most accurate way. Due to that, models generated to provide information about the future state of the road are, usually, incomprehensible to a human operator, making impossible to give him/her an explanation about the causes of the prediction. Given the capacity of rule based systems to explain the reasoning followed to classify a new pattern, the advantages and disadvantages of such approaches are explored in this work. To conduct such task, datasets recorded from the California Department of Transportation are created. A 9-kilometer section of the I5 highway of Sacramento is used for this research. Two different types of datasets are built for the experimentation. One of them contains the entire information recorded. The other one contains with a simplified version of the information, considering only the first, middle and last monitored points of the road. Twelve prediction horizons, from 5 to 60 minutes, were considered for prediction. An experimental comparative study involving 16 state of the art techniques is performed. Techniques tested include those that fall within the categories of Evolutionary Crisp Rule Learning (ECRL) and Evolutionary Fuzzy Rule Learning (EFRL). These methods were selected since they offer to the final user, not only a prediction, but also a legible model about the way in which the decision was taken. Techniques are compared in terms of accuracy and complexity of the models generated.