Heterogeneous kohonen networks

Abstract

A large number of practical problems involves elements that are described as a mixture of qualitative and quantitative information, and whose description is probably incomplete. The self-organizing map is an effective tool for visualization of high-dimensional continuous data. In this work, we extend the network and training algorithm to cope with heterogeneous information, as well as missing values. The classification performance on a collection of benchmarking data sets is compared in different configurations. Various visualization methods are suggested to aid users interpret post-training results. s. © Springer-Verlag Berlin Heidelberg 2001.