A Comparison of Mamdani and Sugeno Fuzzy Inference Systems for Traffic Flow Prediction

Information on the future state of traffic flow provides a solid foundation for the efficient implementation of traffic control and guidance. The prediction approaches based on fuzzy logic theory is of great interests, because the rule-based inference is similar to the way humans process casual relations and fuzzy linguistic variables provide a natural way to deal with uncertainties. This paper presents a comparative study on a set of widely used Mamdani and Sugeno fuzzy inference systems in the application on the short-term prediction for traffic flow based on the historical recordings. To fulfill the comparison, a series of experiments was designed and performed to evaluate prediction performance for each fuzzy inference system in terms of model complexity, execution time, noise resistance, performance consistency, missing data, and multi-stepahead predictability. Before discussing the primary results, a description on the fuzzy inference systems, evaluation factors and criteria was given. The analyses on the experimental results led to several findings which can be referenced when choosing a FIS for traffic flow prediction based on historical recordings.