Comparative research on methods of dimensionality reduction in high-dimension medical data

Abstract

Big Correlation(BC) is a method of retaining effective information and the correlation coefficient between evaluation target and each extracted feature is maximized. According to the experiment results of LLE and PCA in nonlinear not dense dataset with noise, combined with the idea of BC, improved PCA with the biggest correlation is proposed called Big Correlation PCA(BC-PCA) By extracting the principal components that have the biggest correlation coefficient with target of evaluation, BC-PCA algorithm reduces the dimension. Examples of the three algorithms are given on the front dataset, experimental results show the effectiveness of the three algorithms, concluded that LLE algorithm is not absolutely better than linear method for nonlinear dataset I. INTRODUCTIONlong with the rapid development and wide application of computer technology, appeared a large number of high-dimensional unstructured data information. e.g. text data, biological data, and pathological data. In dealing with "dimension disaster" and "dimension gospel" that arise from high-dimensional data, although dimensionality reduction as an effective means has attracted wide attention, also have a lot of research work, classical linear and nonlinear manifold dimensionality reduction algorithm are very effective only in the ideal artificially dataset, the performance is not good for not artificially dataset [1]. In order to confirm the front conclusion, so according to the characteristics of real Carcinomaofrectum Dataset and the purpose of evaluation, in this paper, examples of LLE, PCA, and BC-PCA algorithm are given on Carcinomaofrectum Dataset, then analyzed and compared three algorithms according to the experimental results in BP Neural Networks.