A new iterative and interactive algorithm called CSN (Classification
by Successive Neighborhood) to be used in a complex descriptive objects
identification approach is presented. Complex objects are those designed
by experts within a knowledge base to describe taxa (monography species)
and also real organisms (collection specimens). The algorithm consists of
neighborhoods computations from an incremental basis of characters using
a dissimilarity function which takes into account structures and values of the
objects. A discriminant power function is combined with domain knowledge on
the features set at each iteration. It is shown that CSN consistently outperforms
methods such as identification trees and simplifies interactive classification
processes comparatively to search for K-Nearest-Neighbors method.
