Building classification rules for case-based classifier using fuzzy sets and formal concept analysis

The focus of this paper is a construction of better knowledge base in case-based classifier system. Our knowledge base structure is based on concept lattice where rules are built from its subconcept-superconcept relation. Since the lattice can only be constructed from inputs with binary attributes, descriptive and numeric attributes must be transformed to binary attributes. In this paper, we propose the transformation of numeric attributes to descriptive attributes using fuzzy set theory. We experiment on benchmark data sets, Car and Iris, to determine the performance in term of number of rules used and classification precision. The results show that trend of accuracy is proportional to the size of learning inputs. The number of rules used is relatively small compared with size of training data. Our case-based classifier produces very promising results in practice and can classify the new problem more accurate than traditional classifiers.