Integration of Statistical Analyses for Parametrisation of the Fuzzy Pattern Classification

The present work aims at a statistically motivated parameterisation for a fuzzy classification approach. Its key points are the determination of robust parameterisations for the data-driven fuzzy class learning based an statistical analyses as well as the preservation of the interpretability of the fuzzy class models and the classification process. In particular, order statistics and Monte Carlo methods are used to determine distributions and moments of class border parameters. These distributions and moments are further applied to evaluate the robustness of parameters of the current fuzzy classification model and to propose alternative robust parameterisations. The feasibility of the approach is demonstrated in the context of a machine diagnosis application.