ConceptualClusteringofObjectBehaviourintheDynamicVisionSystemBasedonQualitativeSpatio-TemporalModel.docVIP

Conceptual Clustering of Object Behaviour in the Dynamic Vision System Based on Qualitative Spatio -Temporal Model Maja Mateti? Faculty of Philosophy Rijeka University of Rijeka Omladinska 14, 51000 Rijeka, Croatia maja@mapef.pefri.hr Slobodan Ribari? Faculty of Electrical Engineering and Computing University of Zagreb Unska 3, 10000 Zagreb, Croatia slobodan.ribaric@fer.hr Abstract - A class of machine learning methods relevant to data mining and knowledge discovery concerns the problem of building a conceptual classification of a given set of entities. The choice of the attributes in describing a given input has crucial impact on the classes induced by a learner. We investigate the problem of attribute selection in an unsupervised learning task in the dynamic vision system. The input data obtained by tracking of the laboratory animal in two-dimensional scene is transformed on the base of the qualitative spatio-temporal model and background knowledge, resulting with qualitative description of behaviour. The qualitative approach reduces the number of attributes used to describe the training data in order to improve the quality of the concepts produced. Moreover, new attributes are derived as composite functions of the initial attributes. This data preprocessing is the base for conceptual clustering process which is formulated as clustering that exploits background knowledge and facilitate the interpretation task to user. Obtained clusters are the base for the behaviour analysis. I. INTRODUCTION One of the basic tasks of machine learning and data analysis is to automatically uncover the qualitative and quantitative patterns in data sets. When a large database is given, to discover the inherent patterns and regularities becomes a challenging task, especially when no domain knowledge is available or when domain knowledge is too weak. Because of the size of the database, it is almost impossible for a decision maker to manually abstract from it the useful patterns.