The Kodiag System Case-based Diagnosis With Kohonen Networks

Abstract

This paper describes the case-based KoDiag system, a diagnostic tool based on the Kohonen model of artificial neural networks (ANN), extended by modifications to increase storage capacity and processing speed during learning. A new training method is introduced, that leads to clustering in the Kohonen map according to the feature context and improves performance during the diagnosis process when input data is partially not available. Unlike common ANN-approaches to diagnosis, KoDiag contains both a classification and a test selection component. The classi3cation results of KoDiag are compared to a CBR-expert system.