KPCA Feature Extraction Based on CBPSO

How to choose the best or near kernel function to reduce test error rate is the key of KPCA applied to extract nonlinear feature. In this article, on the basis of research of CA, PSO, we propose a programmer flow of CBPSO used for training kernel function and build CBPSO-KPCA. This approach can effectively optimize kernel function. Simulation results show that produces highly competitive results at a relatively low computational cost. Keywords—CBPSO algorithms; KPCA; feature extraction CA(Cultural Algorithm) is considered as a new evolutionary algorithm (1) . It is originated by Reynolds in 1994. The PSO (Particle swarm optimization) is originated by Eberhart (2) . It is begun in the development of bird's hunt behavior. In the article the PSO is combined with the CA model. It is full use of PSO's swift evolution ability and via the inheritance operation in the CA model to increase the variety of the population together; all these are built the CBPSO (Cultural based PSO). The KPCA (Kernel Principle Component Analysis, KPCA) is the technique of input and output characteristic non-linear change (3) , the analysis of in the characteristic space, in order to obtain a optimization characteristic change in the ability of failure test identify. Because of the existence of selecting all kinds of kernel functions and parameter, how to select the optimization of kernel function and parameter in order to reach the optimization test effect, is a problem and unsolved yet. According to the characteristic of kernel function parameter optimization, the article designs a new algorithm, use the CBPSO to train the kernel function, use the CBPSO-KPCA algorithm for feature extraction. The simulation experiment shows that the combine heightens the optimization of kernel function effectively, it overcome the difficulty of kernel function application in some extents.