Technical Report, SFB 475: Komplexitätsreduktion in Multivariaten Datenstrukturen, Universität Dortmund 2002,10
The advanced maximum-linkage-algorithm (AMLA) is a derivative of the maximum-linkage-algorithm (MLA) given by Zerbst (2001). AMLA produces clusterings which have a good separation between the built classes. To reach the separation, centroids were calculated, which are the basis for classification. The separation between the classes is considered large, if this is also true for the so far calculated centroids. The selection of centroids having a large distance to each other is also guaranteed using the MLA. AMLA is improved in this way that the underlying frequency structure of the data can be adapted individually to the problem under consideration. Therefore an additional parameter p∈[0,1] is introduced. The parameter gives the degree to which the frequency structure is regarded.