A MODIFIED CASE-BASED REASONING METHOD BASED ON THE ROUGH SET THEORY

Context. Knowledge bases are the main element of artificial intelligence systems. They are formed on the basis of two generally accepted approaches: the object-oriented approach and object-structural approach. Knowledge structuring through its ordering, classification and typing of selected classes is the main operation that is implemented in both approaches. Quite often there are situations when data or knowledge is not exact and it is impossible to perform their exact classification. These features necessitate the development of new approaches aimed at solving problems of extracting knowledge from large arrays of unordered data, structuring, presenting and analytical processing of inexact knowledge in automated construction of knowledge bases. Objective. The objective of this paper is a research of new approaches for solving problems of representation of knowledge about cases in intellectual decision support systems. Method. An approach aimed at modifying Case-Based Reasoning method on the basis of Rough Set Approach has been proposed in this paper. The proposed method forms a partition of cases to determine the degree of their belonging to the goal classes using upper and lower approximations of goal classes, considering the relative importance of classification attributes and formed equivalence classes. Results. The proposed modification of Case-Based Reasoning method allows extracting knowledge about cases from arrays of unordered data with the purpose of the case base construction, and handling the inconsistent (in cases where for the same values of attributes cases belong to different classes), and incomplete (in cases where the values of some attributes or information of the case belonging to the given class is missing or unreliable) information about cases. Conclusions. The proposed method of representation knowledge about cases, their adaptation and subsequent search in the case base formed under uncertainty and existence of inexact, rough, inconsistent initial data constitutes a theoretical basis for constructing intellectual decision support systems.