Process identification and quality control with evolutionary optimized RBF classifiers

Data mining algorithms are needed in many industrial applications in addition to conventional algorithm toolboxes of process and control engineers. Two key tasks that must often be mastered in applications dealing with classification problems are the selection of input features for a classifier (attributes) from a given, often large set of possible features and the optimization of the classifier's structure with respect to the selected input features. These two problems - feature and model selection - should be addressed simultaneously to achieve the best classification results. This article describes an evolutionary algorithm that performs feature and model selection for classifiers based on radial basis function networks. The advantages of this approach are set out by means of two industrial application examples in the areas of process identification and quality control.