Extraction of representative learning set from measured geospatial data Béla Paláncz 1 , Lajos Völgyesi 2 , Piroska Zaletnyik 2 , Levente Kovács 3 1 Department of Photogrammetry and Geoinformatics, Faculty of Civil Engineering, Budapest University of Technology and Economics, 1111 Budapest, Műegyetem rkp. 3, palancz@epito.bme.hu 2 Department of Geodesy and Surveying, Faculty of Civil Engineering, Budapest University of Technology and Economics, volgyesi@eik.bme.hu 3 Department of Control Engineering and Information Technology, Faculty of Electrical Engineering and Informatics, Budapest University of Technology and Economics, lkovacs@iit.bme.hu Abstract: The efficiency of the application of soft computing methods like Artificial Neural Networks (ANN) or Support Vector Machines (SVM) depends considerably on the representativeness of the learning sample set employed for training the model. In this study a simple method based on the Coefficient of Representativity (CR) is proposed for extracting representative learning set from measured geospatial data. The method eliminating successively the sample points having low CR value from the dataset is implemented in Mathematica and its application is illlustrated by the data preparation for the correction model of the Hungarian gravimetrical geoid based on current GPS measurements. Keywords: machine learing, representativness of data, geospatial data. 1 Introduction During the last decade, machine learning algorithms, such as artificial neural networks (ANN) and support vectors machines (SVM) have extensively used for wide range of applications. They have been applied for classification, regression, feature extraction, data prediction and spatial data analysis. To ensure generalization properties of machine learning methods like artificial neural networks and support vector machines, the set of measured data should be split into learning and testing sets, [1]. The question is how to divide the measured sample set into these three sets in order to extract the most information as it is