Application of manifold learning and nonlinear dimensionality reduction in private preserving