Kernel principal component analysis in the application of geochemical comprehensive feature extraction

Abstract

Principal component analysis is a kind of effective method of extracting comprehensive geochemical data feature. By constructing a new comprehensive variable to instead of the original variables, the new can effectively reflect the compositive information of original variables; it also could indicate the pargenetic assemblage and genetic relationship of exploration geochemistry. But it is based on the hypothesis premise of the normal (liner) distribution of the sample data. However, the complexity of geological systems and multiple stage mineralization stage often lead to the nonlinear distribution of multivariate geochemical data. Therefore, compared with the traditional principal component analysis, the nonlinear principal component analysis is more suitable for extracting of the multivariate geochemical data. This paper introduces the principal component analysis basing on kernel function. With the help of a "nuclear techniques", implicitly map the input space to a nonlinear characteristics space. In this space, we carry out principal component analysis of geochemical data. The algorithm is in line with the exploration geochemistry data features. Through the experimental analysis of Tibet Daewoo stream sediment data, the principal components analysis based on kernel function is compared with the conventional PCA can better complete the comprehensive exploration geochemistry data feature extraction.