Improved Fuzzy based Kernel PCA

In the non-linear principal component analysis processing, the kernel-based nonlinear dimensionality reduction technique KPCA is great sensitive to large deviation samples, while RKF-PCA is non-convergence due to improper parameter selection. A Improved Fuzzy Kernel Principal Component Analysis (IFKPCA) algorithm, which managed through weighting the sample points by a membership function included fuzzy parameters C, is introduced based on fuzzy theory. Various distribution functions, including large deviation samples or not, are tested using conventional KPCA, RKF-PCA and IFKPCA separately. The results show that, IFKPCA weakened the impact of the large deviation samples, and avoided the non-convergence problem, cased by improper parameter selection. Besides, IFKPCA is robust, and the selection of the weight coefficient parameters of IFKPCA is also convenient. So IFKPCA is a good solution to the samples sensitive problem of KPCA.