Hierarchical Inference
Edwin Diday · 1984
It often happens in practice, that a user wishing to make a hierarchical classification, does not know which of the panoply of dissimilarity indice will be the best one for his data. It can also happen that none of these indices satisfies the data that he must deal with. If the user has at the outset some ideas on the classification that he wishes to obtain, our approach permits to induce an aggregation indice, from knowledge acquiring on a learning set, given by the user. Constraint have been defined in the search of the indice, to ensure that no inversion takes place. A nearest neighbours algorithm, with constraints of validity, is used to induce the final hierarchy. An application to pattern recognition is proposed, which has permitted to induce some aggregation indices adapted to particular two-dimensionnal repartition of points.