Support vector machines is one of the most popular binary classification methods in machine learning. One of the main problems of SVMs is its sensitivity to noise and outliers. Various methods have been proposed to overcome this problem. Fuzzy SVMs (FSVMs) main purpose is to consider the constant C in SVM (which is a penalty term to determine the trade-off between margin maximization and training error minimization) as a vector that will permit to some samples to have different amount of logical movement, so that it has been placed at an appropriate location in the feature space. Our proposed method is to use locally linear embedding (LLE) neighborhood concept to obtain a good penalty vector which allows noisy data to be placed at an appropriate location. Experiments indicate the superiority of the robustness of our proposed method dealing with noise impregnated data comparing with traditional SVM.

Keywords