Characteristics Of Medical Data Based On Pca Extraction Method Research And Application

In processing of high dimension data, if we ignore the problem of dimension reduction, the computation will be so heavy that it's difficult to get valuable information. Feature extraction is an important preprocessing technology in data mining. Feature extraction is the operation of acquiring available features for classing from original characteristics of the high dimensional objects in order to represent these objects properly, through which we could reduce feature space. In this paper, on the background of Natural Science Foundation of Hunan Province of China under Grant No.06JJ50143, laser-induced fluorescence spectra feature extraction technology is studied systemically.In this paper, we analyze and compare two representative feature extraction methods and their advantages and disadvantages. Then, we propose two feature extraction methods.Firstly, from statistics analysis perspective, a novel fluorescence spectra feature extraction algorithm named PCA_AFLDA is presented. It overcomes bad robustness and unsupervised learning in PCA. Also, it solves small sample problem and rank- limited problem in FLDA.Secondly, from knowledge discovering perspective, based on similitude entropy, we import Rough Set theory and propose a Rough PCA(RPCA) approach. It doesn't need any prior knowledge and can analyze and process knowledge narrowly.The experimental results show that RPCA solves the problem of the number of samples and uncertainty of information in PCA_AFLDA. Also,it has higher discriminating accuracy, sensitivity and specificity than PCA_AFLDA.