We have developed new methods and tools for subgroup discovery, SD, CN2-SD and APRIORI-SD, based on algorithms for learning classification and association rules. The main methodological achievement are new machine learning algorithms for discovering subgroups, new methods for evaluating the quality of subgroups and the theoretical analysis with a collection of advice for using these methods in practice. We have applied the algorithms in the area of marketing, in biomedicine for analyzing microarrays and in medicine for determining risk groups for coronary disease.
COBISS.SI-ID: 18092839
We have developed relational subgroup discovery (RSD) data mining method for discovering regularities in multi-relational databases and applied this method in analyzing the mutagenicity of molecules, analyzing telecommunications data and in virtual design of automobile parts. The RSD method was extended with the ability to use domain knowledge in the form of ontologies (eg. the GO gene ontology) and used to analyze microarrays. The importance of this achievement for bioinformatics is in efficient inclusion of biological knowledge, publicly available on the Web, into automatic learning process.
COBISS.SI-ID: 19724583
The development of equation discovery systems is a breakthrough research achievement as our systems can discover algebraic, ordinary and partial differential equations; they can also take into account existing domain knowledge. We have introduced a formalism for representing domain knowledge for modelling dynamic systems, based on the notion of processes, developed methods for using such domain knowledge and used them to model aquatic ecosystems. We have co-edited the book "Computational Discovery of Scientific Knowledge" which represents a reference state-of-the-art survey in this area.
COBISS.SI-ID: 21312295
We have developed new methods of knowledge discovery from text: the system OntoGen for (semi)automatic ontology building from a corpus of documents, a method for building ontologies from social networks, a method of efficient insertion of examples into ontologies, a method of handling unmarked data with active learning methods, a method of dealing with context in ontology learning and a method of vizualising text corpora. The methods are included in our publicly available library of programs TextGarden. OntoGen received the best system prize at the 3rd European Semantic Web Conference.
COBISS.SI-ID: 19498279
We have developed methods and tools for decision support, based on qualitative multi-parametric modeling. Our core method DEX was extended with new features such as the work with proper hierarchies. The book Decision Making and Models was published, describing many years of our experience in decision support. We developed methods for revising decision models and implemented them in the system proDEX. Our methodology was used in practice in analysis of effects of using genetically modified crops, in traffic management, for selecting the design of Slovene coins and for managing electronic waste.
COBISS.SI-ID: 20073767