关键词:正则化; Fisher判别分析; 核方法; 凸优化; 支持向量机
Kernel form of regularized FDA and comparison study with SVM
YU Chun-mei, PAN Quan, CHENG Yong-mei, ZHANG Hong-cai
(College of Automation, Northwestern Polytechnical University, Xi’an 710072, China)
Abstract:Whereas small sample size (3S) problem will be arose in both FDA and KFDA. Regularized FDA is an effective solution for this problem. To study the comparison of regularized FDA and support vector machine (SVM), this paper derived a novel kernel form of regularized FDA, which transfered optimization problem with constraint to optimization problem in dual space. Obtained the kernel form which similar to SVM and gave the links with SVM. Simulation results for Tenessee-Eastman(TE) process show that regularized KFDA get better diagnosis effects than least squares SVM(LS-SVM). ......