MPOC: an agglomerative algorithm for multicriteria partially ordered clustering (Q385464): Difference between revisions

The paper proposes a multi-criteria clustering method, for problems where alternatives have to be clustered into classes with preference relations between classes. The authors first discuss a partial order \(\tau\) between classes derived from a given outranking relation \(S\) on alternatives. They first introduce a strong \(S\tau\)-consistency principle before discussing relaxed forms of this consistency, such as \(S\)-consistency, \(\tau\)-consistency, and semi-strong \(SS\tau\)-consistency. With these definitions they proceed to define an outranking degree and outranking relation between classes. They define three indicators \(v_P, v_I\) and \(v_R\) measuring how much a partition respects the principles of preference, indifference, and incompatability between elements. The proposed algorithm follows the agglomerative hierarchical structure of clustering algorithms. The authors then propose an algorithm that restores transitivity of the resulting partition. The method is illustrated on a data set for sorting stocks listed on the Athens Stock Exchange.