Method of extraction of causal relationships with optimized fact bases

In this paper a method of extraction of causal relationships on the set of fact bases obtained by learning on data from weakly formalized subject area is proposed and studied. Fact bases are built for the target properties of each object class. Learning is carried out with the use of co-evolutionary genetic algorithm, which reduces the initial feature space. The formed class descriptions obtained by means of the first phase of the JSM-method are used for search of causal relationships for all target properties. The proposed method is suitable both for the analysis of a small amount of data and for work with sets of incomplete data of the big size. A number of model experiments with use of base of medical data MIMIC II are carried out.