An indiscernibility-based clustering method with iterative refinement of equivalence relations

Abstract

In this paper, we present an indiscernibility-based clustering method that can handle relative proximity. The main benefit of this method is that it can be applied to proximity measures that do not satisfy the triangular inequality. Additionally, it may be used with a proximity matrix - thus it does not require direct access to the original data values. In the experiments we demonstrate, with the use of partially mutated proximity matrices, that this method produces good clusters even when the employed proximity does not satisfy the triangular inequality.