Classification Criterion Based Neighborhood Optimization Method on Laplacian Eigenmaps

Abstract

Manifold learning algorithms are nonlinear dimensionality reduction algorithms rising in recent years. Laplacian Eigenmaps is a typical manifold learning algorithms. Aim to the difficulty of selecting neighborhood parameter on the algorithm, a neighborhood parameter optimization method based on classification criterion is proposed in the paper. From the point of the classification performance, the classification criterion function is constructed to reflect the distance of within-class and between-class. The optimization of the neighborhood is obtained according to the minimum of the criterion function. The experimental results on IRIS validate the optimization of the neighborhood and the effectiveness of feature classification.