Distance weighted fuzzy k-NN SVM

By the inspiration of the Gaussian noise reduction, the study extends a local-global hybrid learning system to make a classification robust. The system formed by a two-stage scheme with a preprocessor of nonparametric density estimator to evaluate locally the contaminant level of the degenerated training data, and a postprocessor of fuzzy support vector machine to learn globally an optimized classifier. As expected, the fuzzified classifier could achieve a better robustness. The improvement is indeed carried out by the extension which extends the nonparametric density estimators to a more general distance-weighted versions. Both the theoretical elaboration of the hybrid system and corresponding simulations are included in this paper.