Finding Relevant Process Characteristics with a Method for Data-Based Complexity Reduction

In this paper we present a fuzzy method that allows a reduction in the number of relevant input variables and, therefore, in the complexity of a fuzzy system. Input variables are only kept if they affect the output variable and are not correlated with any other kept input variable. The efficiency of the method is demonstrated by using it to find relevant process characteristics in the examples of a fuzzy classifier for quality control in the car industry and a load predictor used in power station management.