Flatness and continuity in the percentile clustering model (Q811455): Difference between revisions

Cluster analysis is a process of classifying objects into subsets with similar properties. A cluster method can be interpreted as a transformation of dissimilarity coefficients characteristic for a given set of objects. In the present paper, dissimilarities are measured in a given lattice L and cluster methods are transformations on sets of residuated maps. An isotonic function \(f: P\to Q\) is residuated if there exists an isotonic function \(f^+: Q\to P\) such that \(f^+f(p)\geq p\) and \(ff^+(q)\geq q\). The system of such maps is denoted Res(P,Q). A transformation \(\Gamma\) from Res(M,L) to Res(N,L) is called an L-cluster method. In the case when L is the lattice of percentile functions \(\Gamma\) is called a percentile cluster method. Important properties of cluster methods are flatness, monotonicity, equivariance and connectivity. The main result of the paper is the fact that induced flat percentile cluster methods are monotone, equivariant and continuous.