Support Vectors Classification Method Based on Matrix Exponent Boundary Fisher Projection

摘要

Boundary Fisher analysis is a supervised dimension reduction method, the original data characteristics are preserved after dimension reduction, but singularity of matrix is involved in data optimization. By boundary Fisher thought, support vectors classification method based on matrix exponent boundary Fisher projection is proposed. In the process of Fisher optimization, the inverse matrix is transformed into the matrix exponent inverse matrix, and solve singularity problem of Fisher optimization, then obtain the optical projection matrix, the original training samples are projected to the low dimension space, which are used to train support vector machine(SVM). Experiments on two artificial data sets and UCI standard data set show that the proposed method can solve singularity problem, especially in the small samples.