Feature Selection Methods for Softcomputing Classification

Feature selection and feature creating are two of the most important and difficult tasks in the field of pattern recognition. It involves determining a good feature subset given a set of candidate features. The acoustic analysis of vibration signals in the time and frequency domain frequently generates a large number of features and makes a reduction of dimensionality necessary. The present method is an approach to improve pattern classifier performance using a feature selection process. For this task the two parts feature selection and the inherent classification step are combined. The results obtained from an application for the automatic detection of different surface structures indicates the usefulness of the approach. The algorithm is realized as the Plug-In FeatureSelector for the software tool Data Engine .