A goal-related distance function on fuzzy partitions and its use in nearest neighbour classification

Abstract

Fuzzy similarity measures have been widely studied but their relationship with classification tasks have hardly been emphasized. Besides, what they just consider is the superficial difference between the objects to be compared without a deeper understanding of the domain. Intuition suggests that the similarity of objects needs to be modified in accordance to the manner in which they are classified. From the cognitive point of view it is clear that human beings modify their general knowledge as they achieve new experiences. This paper proposes a new distance measure based in the concept of fuzzy association between a predictive (fuzzy) value and the class value in a given memory of instances. This type of goal related similarity measure is new in the fuzzy context.