The problem analyzed in the thesis is biomedical diagnostics. Its specifics is several thousand biological indicators of a patient status (genes, proteins and antibodies) that have to be analyzed simultaneously in order to find disease markers. This problem is formalized in the thesis as data mining classification task, in which patient status is described by vectors made up of biological indicator values and the diagnosis of the patient is the class label. A methodology is developed to solve the defined task, using two methods developed for the solution of this task – class decomposition and a hybrid classification method that is based on genetic algorithms and decision tree classifier ensembles. Class decomposition allows improving classification accuracy by describing the inner structure properties of the data and using the description in classification. The classification method that is based on genetic algorithms and decision tree classifier ensembles and that uses Random subspace method allows finding quasi-optimal and easily interpretable classifier ensembles that consist only of the most informative attributes and their relationships. The thesis is arranged so that confirming the initial hypotheses step-by-step it proves the efficacy of the developed methodology. As a result, the use of a smaller biomarker panel that is acquired due to the built-in feature selection of the developed method is justified, the usefulness of class decomposition application is proved, the accuracy of the developed classification method is confirmed and the advantages of using the developed methodology for the analysis of biomedical data are shown.