Pattern recognition is a central task in a variety of applications ranging from engineering to financial analysis. For this reason, it is very important to develop efficient pattern recognition systems that help to make decisions automatically and reliably. This thesis describes the implementation of pattern recognition systems based on computational intelligence approaches (in particular support vector machines and radial basis function neural networks). To improve the efficacy of these systems, a notion of prototypes stability analysis is introduced and fuzzy kernels are developed. Prototypes stability analysis is used for determining the adequate number of prototypes, clusters, or information granules to be used in classifiers design. Fuzzy kennels, optimized using genetic algorithms, allow for the incorporation of fuzzy set methods into the support vector machine approach. To demonstrate the applicability of these pattern recognition systems, detailed numerical experiments with synthetic and real-world data sets are included and analyzed. |