ASSESSMENT OF MODELS FOR PEDESTRIAN DYNAMICS WITH FUNCTIONAL PRINCIPAL COMPONENT ANALYSIS TIM ENSSLEN, HANNO GOTTSCHALK AND MOHAMED SAADI Department of Mathematics and Computer Science, Bergische Universit¨at Wuppertal, 42119 Wuppertal, Germany, {tim.ensslen,hanno.gottschalk,saadi}@uni-wuppertal.de MOHCINE CHRAIBI AND ARMIN SEYFRIED J¨ ulich Supercomputing Centre, Forschungszentrum J¨ ulich GmbH, 52428 J¨ ulich, Germany and Department of Civil Engineering, Bergische Universit¨ at Wuppertal, Pauluskirche 7, 42285 Wuppertal, Germany {m.chraibi,a.seyfried}@fz-juelich.de Many agent based simulation approaches have been proposed for pedestrian flow. As such models are applied e.g. in evacuation studies, the quality and reliability of such models is of vital interest. Pedestrian trajectories are functional data and thus functional principal component analysis is a natural tool to asses the quality of pedestrian flow models beyond average properties. In this article we conduct functional PCA for the trajectories of pedestrians passing through a bottleneck. We benchmark two agent based models of pedestrian flow against the experimental data using PCA average and stochastic features. Functional PCA proves to be an efficient tool to detect deviation between simulation and experiment and to asses quality of pedestrian models. PACS numbers: 89.75-k Complex Systems, 50.40-a Stochastic Models Keywords: pedestrian dynamics; statistical analysis; comparison with experiment; functional PCA; model quality arXiv:1502.00528v1 [physics.soc-ph] 2 Feb 2015