Hierarchical fuzzy rule-based system optimized with genetic algorithms for short term traffic congestion prediction

• Data about 87 loop detectors in a real highway during a month is taken.
• 3 Datasets about congestion in the next 5, 15 and 30 min are defined.
• A codification for hierarchical fuzzy rule-based system is stated.
• A genetic algorithm to optimize such hierarchical systems is implemented.
• Results are compared with other classifiers in terms of accuracy and simplicity.

Taking practical and effective traffic prediction and control measures to ease highway traffic congestion is a significant issue in the research field of Intelligent Transportation Systems (ITS). This paper develops a Hierarchical Fuzzy Rule-Based System (HFRBS) optimized by Genetic Algorithms (GAs) to develop an accurate and robust traffic congestion prediction system employing a large number of input variables. The proposed system reduces the size of the involved input variables and rule base while maintaining a high degree of accuracy. To achieve this, a hierarchical structure composed of FRBSs is optimized by a Steady-State GA, which combines variable selection and ranking, lateral tuning of the membership functions, and optimization of the rule base. We test the capability of the proposed approach on short term traffic congestion problems, as well as on benchmark datasets, and compare the outcomes with representative algorithms from the literature in inferring fuzzy rules, confirming the effectiveness of the proposed approach.