Zaiko T. Methods of association rule extraction in intelligent diagnosis systems

The purpose of the thesis - the development of methods for association rules extraction to improve the generalization and interpretability levels of synthesized diagnostic models. The object of research - the process of association rule extraction. The subject of research - the methods of association rule extraction in the intelligent systems. New methods and software that allow to mine the association rules to improve the generalization and interpretability levels of synthesized diagnostic models are developed. The method of numerical association rules is proposed, which taking into account the individual importance of features, using the criteria for evaluation of indirect associations. The method of factor analysis based on the association rules, which involves extracting rules from a given database of transaction, resulting in the synthesis of data is performed, and therefore, the exclusion from further consideration redundant features, which can reduce the search space and time performance of factor analysis is developed. The method of the dimension reduction of training samples based on the association rules is proposed. The method for the neuro-fuzzy networks synthesis based on the association rules is developed. Numerical experiments for solving of practical problems of automation of technical and medical diagnostics and study of the properties and characteristics of methods of association rules mining are provided. The obtaіned rеsults are enhаnced to the practіcal realizatіon and іntroduced in enterprіses and organіzations.