Abstract—With new services provided by airlines and travel agencies passengers gained more flexibility, but also their visibility for airports and airlines was decreased. We describe the profiling component of a movement forecast system to increase passenger transparency. Therefore we present three approaches to learning classifiers and how they can be used for passenger classification. It is shown how the choice of attributes considered in classification influences classifier performance. This is used as a comparison criterion for learning algorithms. Index Terms—accuracy estimation, comparing classifiers, feature subset selection, passenger classification I. INTRODUCTION The constantly increasing number of air traffic passengers over the last years requires either the expansion of existing infrastructures like terminal buildings or the construction of new airports. Modern terminal buildings can be designed with a layout that allows different streams of passengers being separated from each other in compliance with regulations for passenger flows like domestic, Schengen, international, arriving, departure, or transit. However, existing terminals are restricted in their infrastructure and have to find a possible integration of regulations and control for the passenger flows. In addition, passengers get lost, and suffer from raised stress levels. They might cause the delay of flights due to different passenger handling philosophies – from past time periods – as well as regulation by authorities varying from country to country. Airlines and travel agencies continuously improve their extensive services to manage and simplify the passenger’s travel with respect to flexibility. For example, passengers can check in at home to avoid long queuing lines at the airport – if they travel without check-in luggage (common for domestic or short range business traveler). Unfortunately, these passengers are not visible for the airlines until the moment where they pass the departure gate. Note that passenger pass through several security checks and border controls without any notification to the airlines. In this publication, we discuss some aspects of the Manuscript received December 8, 2008. S. Richter is with EADS Innovation Works of the EADS Deutschland GmbH, 21129 Hamburg, Germany (phone: +49-40-74381510; fax: +49-40-74381517; e-mail: stefan.richter@eads.net) C. Ortmann is with EADS Innovation Works of the EADS Deutschland GmbH, 21129 Hamburg, Germany (e-mail: christoph.ortmann@chortmann.de) T. Reiners is with Institute of Information Systems of the University of Hamburg, Germany (e-mail: reiners@econ.uni-hamburg.de) methodology for a system to support airlines and airports with respect to the transparency of passengers, i.e. to detect late- or no-show passengers as early as possible. II. SYSTEM OVERVIEW We developed an airport movement forecast system to detect late- or no-show passengers in an early stage and to predict the routes of passengers – based on their characteristics – already being present at the airport; with calculations being done in or close to real-time. In addition, the system can provide optimized routes for the single passenger to guide him smoothly and without stress throughout the terminal. That is, passengers can ask for guidance to so-called Points of Interests (POI) and the system will determine the (best) route in accordance to feasibility checks. These checks include estimations if the passenger can visit his POIs and still arrive at the gate in time. Furthermore, the system can determine if the POI is out of scope for the passenger, e.g. a domestic traveler cannot access the duty free area. There are other minor checks to assure plausible results of the system but these are not further discussed in this paper. For the system a modular approach is used to provide flexibility for different operating systems and hardware architectures, or a later system optimization being described in Section VII. The modular architecture also supports extensibility, scalability, and refactoring of modules to prevent large or unstructured systems. All other components are linked to a centralized database system. The database stores third party data like up-to-date passenger data from airlines, travel agencies, and tracking data from the airport and ground handlers. Other components of the overall system are modules dealing with, e.g., communication, calculation, or classification. In the current development stage, the route calculation module cover the complete route prediction calculation, see [1]. It was planned from the beginning that this is an intermediate step as the prediction of reliable movement information for each passenger had to be improved using more sophisticated algorithms. Nevertheless, the algorithms were chosen in a way that they can later be modified by another system component. In this paper we introduce passenger classifier, which are used for calculations to forecast the routes according to the specific characteristics of individual passengers. III. SYSTEM IN- AND OUTPUT The prediction of routes requires a certain set of passenger attributes that are obtained from third party applications. The results of our calculations can be redistributed to third party Passenger Classification for an Airport Movement Forecast System Stefan Richter, Christoph Ortmann, Torsten Reiners Proceedings of the International MultiConference of Engineers and Computer Scientists 2009 Vol I IMECS 2009, March 18 - 20, 2009, Hong Kong ISBN: 978-988-17012-2-0 IMECS 2009