V.012 SHORT TERM PREDICTION OF HIGHWAY TRAVEL TIME USING DATA MINING AND NEURO-FUZZY METHODS 1 David Coufal 2 Institute of Computer Science Academy of Sciences of the Czech Republic Pod Vodarenskou vezi 2, 182 07 Prague 8, Czech Republic Esko Turunen 34 Tampere University of Technology P.O. Box 692, 33101 Tampere, Finland Abstract We show that prediction of travel time on a 28-km long highway section based on on-line travel time measurements with video is practicable by data mining and neuro-fuzzy methods. We introduce two new prediction models. The first one is a result of GUHA style data mining analysis and Total Fuzzy Similarity method, and the second one is a hierarchical model based on neuro-fuzzy modelling. Comparing results with the existing Traficon model, both new models improve the travel time prediction. The results obtained by the new methods are comparable to MLP neural network model, too. Key words: fuzzy logic, neural networks, data mining. 1. Introduction The aim of this study is to show that short-term travel time prediction presented in [3] can be carried out by data mining and neuro-fuzzy methods, too, and that results are comparable. Research [3] was carried out on main road 4 between points A (Lahti) and D (Heinola) in Southern Finland. According to [3], the average daily summertime traffic on this 28-kilometer section is about 15100 vehicles per day, in particular, the traffic volumes are high during summer weekends. The study section AD is divided into three sub-sections AB, BC and CD with camera stations approximately equally distributed over link AD length and equipped with an automatic travel time monitoring system. The system is based on an artificial vision and neural network application, which automatically reads license plates. Moreover, there is an inductive loop detector on station C gathering information on traffic volumes and point speeds. A variable message sign (VMS) at point A gives upper and lower bounds of a estimations about the travel to the point D. In an unpublished preliminary study of the problem done by Laura Lanne, the estimation categories are below 25 min, 25-30 min, 30-40 min, 40-50 min and above 50 min. In [3], travel time from point A to point D is regarded as congested if it is above 25 min. In [3], travel time prediction p is regarded as acceptable if the real travel time lies in the interval [0.9*p, 1.1*p], i.e., it is used ±10% marginal error based correctness of results. 1 This research is part of research project COST Action 274 [TARSKI] 2 Supported by grant OC 274.001 of Ministry of Education, Youth and Sports of the Czech Republic 3 Supported by grant 2000085 of Finnish Academy 4 Correspondence via e-mail esko.turunen@tut.fi