Fuzzy k-NN SVM

A fuzzy support vector machine emphasized the noise contamination locality in its first filtering stage is proposed. As a consequence to assign locally fuzzy memberships to the learning samples in the preprocessing filtering stage, the locality enhances the support vector machine, which is originally devised to learn a classifier with the global quadratic optimization, to compromisingly adapt to the individual attitude of the learning data. The paper employed a fuzzy k-NN rule as the preprocessor. The k-NN approach is advantageous to with its nonparametric nature, learning directly from the given prototypes without additional complex computation, is really appropriate for the local-global combination. By unraveling the individual attitude in the contaminated mess of the dataset as a fuzzy membership, an underlying fuzzy support vector machine is thus applied to finish the model. The model, originated as a variety of the fuzzy support vector machine, not only shares the merits of its crucial robustness which inspired by the global optimization, but also exhibits its capability in keeping the room for the learning samples in their representation of local confidence.