Rough Set Approach to CBR

Abstract

We discuss how Case Based Reasoning (CBR) (see e.g. [1], [4]) philosophy of adaptation of some known situations to new similar ones can be realized in rough set framework [5] for complex hierarchical objects.

We discuss how various problems can be represented by means of complex objects described by hierarchical attributes, and how to use similarity between them for predicting the relevant algorithms corresponding to these objects. The complex object attributes are of different types: basic attributes related to problem definition (e.g. features of object parts), attributes reflecting some additional characteristic of problem (e.g. features of more complex objects inferred from properties of their parts and their relations), and attributes representing algorithm structures (e.g. order and/or properties of operations used to solve the given problem). We show how to define these particular attributes sets, and how to recognize the similarity of objects in order to transform algorithms corresponding to these objects to a new algorithm relevant for the new incompletely defined object [1],[4].

Object similarity is defined on several levels; basic attribute recognition level, characteristic attribute recognition level and algorithm operation recognition level. Dependencies between attributes are used to link different levels. These dependencies can be extracted from data tables specifying the links.

We discuss how to classify new objects, and how to synthetize algorithm for such new object, on the basis of algorithms corresponding to similar objects. The main problem is the generation of rules enabling to create operation sequences for a new algorithm. These rules are generated using rough set approach [5].