The results of this research program belong to the area artificial intelligence, more specifically machine learning, intelligent data analysis, automated modelling, and qualitative reasoning. Some results are of basic scientific nature, and some are applications oriented. The third type of results are software tools that support the application of research results. The list of results below, which also mentions key publications, is organised according to this division. - Method of machine learning based on function decomposition, implemented in the program HINT. HINT is an approach to constructive induction by building a hierarchy of concepts. In domains with useful intermediate concepts, HINT achieves spectacular accuracy improvements compared to traditional machine learning methods (http://magix.fri.uni-lj.si/hint; Zupan et al., Artificial Intelligence Journal, 1999; Bohanec and Zupan, Decision Support Systems, 2003) - A new approach to machine learning - learning of qualitative trees, implemented in the system QUIN. The method enables qualitative identification of a system from measured, possibly noisy data. QUIN was instrumental in the participation of this reasarch group in the European project Clockwork (Šuc, monograph, IOS Press, 2003; Bratko and Šuc, AAAI Magazine, 2003). - A new approach to the learning from numerical data, called Q2 learning (qualitatively faithful quantitative learning), which combines qualitative tree learning with the so-called qualitative-to-quantitative transformation (Q2Q). The approch was applied to the optimisation of a complex car model designed by INTEC (Šuc, Vladušič and Bratko, Artificial Intelligence Journal, 2004). - Study of comprehensibility of decision trees with respect to the attribute evaluation method, and the interpretation of evaluations by methods RelieF and RReliefF, which were developed to assess the depenedency between attributes in regression and classification (Robnik and Kononenko, Machine Learning Journal, 2003). - Development and analysis of a general transduction method for the estimation of the reliability of predictions (Kukar and Kononenko, 2002) - An approach to survival analysis (Zupan et al., Artificial Intelligence in Medicine, 2001). - The GenePath approach to the discovery of regulatory networks from genetic data. The approach is based on abductive resoning, supports explanation and proposes new genetic experiments (http://genepath.org; Zupan et al., Bioinformatics, 2003; Artificial Intelligence in Medicine, 2003) - A two-stage method for the synthesis of system control, based on machine learning and qualitative modelling. This approach gave excellent results in skill reconstruction (behavioural cloning) and identification of control systems (Šuc and Bratko, IEEE Trans., 2000; ETAI Trans. 1999 in 2000). - Implementation of Image Processor for the parameterisation of textures, GDV Assistant for analysis of gas ionisation images, and a method for describing texture images with association rules (Kononenko et al., 2003). - Medical and other applications of machine learning: prognosis in wund healing (Cukjati et al. 2001), diagnosis of heart diseases (Kukar et al. 1999), prognosis in hip surgery (Zupan et al. 2003), prognosis in breats cancer; image analysis in detection of wine and apple plant disease (Sadikov et al. 2003), detection of pre-competition stress in sports (Kononenko and Osredkar 2002), analysis of biophysical and psychological parameters with respect to relaxation method (Trampuz et al. 2002), detection in change in state of conciousness (Kononenko et al. 2003) - Development of the technology of DecisionsandHand and implementation of decision shells on web pages and palm computers. This technology which supports the application of decision models developed by means of machine learning was tested in several medical domains (http://magix.fri.uni-lj.si/palm).