7 th Symposium of the European Association for Research in Transportation conference – hEART 2018 5-7 September 2018, Athens, Greece Comparing theoretical pedestrian simulation models with data driven techniques George Kouskoulis a , Constantinos Antoniou b , Ioanna Spyropoulou c a Ph.D. Candidate, School of Rural and Surveying Engineering, National Technical University of Athens, Iroon Polytechneiou Str. 9, Zographos, Athens, 157 80, Greece b Professor, Chair of Transportation Systems Engineering, Technical University of Munich, Arcisstraße 21, Munich, D-80333, Germany c Assistant Professor, School of Rural and Surveying Engineering, National Technical University of Athens, Iroon Polytechneiou Str. 9, Zographos, Athens, 157 80, Greece Keywords: Pedestrian modeling, data driven, loess, social force model, data noise reduction 1. Introduction Following up on the continuous widespread availability of data and computational advances, data driven modeling has been increasingly gaining researchers’ interest over the last decades (e.g. Antoniou et al., 2013; Papathanasopoulou and Antoniou, 2015). Methods and techniques have been developed on clustering, classifying and regressing data with no necessary explicit a priori knowledge of model parameters’ relationships. These methods are not subject to parametric limitations and are thus wider applicable. On the other hand, theoretical models provide a straight mathematical framework, while relating model parameters based on logical principles. They may be categorized as follows (Kouskoulis and Antoniou, 2017):  Cellular automata  Social force  Lattice gas theory Cellular automata models rely on space discretization, social force on the interactive forces that are acted among moving pedestrians and lattice gas on the drift strength (strength of pedestrian flow). Both cellular automata and lattice gas models compute the possibility a pedestrian’s next step based on the surroundings (obstacles and agents). Thus, the basic principle on modelling pedestrian movement lies on the positions of the adjacent pedestrians/objects and their impact on the examined agents. This is directly represented in the dynamics of the social force models. Even more, social force models outweigh on computational requirements.