A0543
Title: Fuzzy cluster-scaled principal component analysis for high-dimension low-sample data
Authors: Mika Sato-Ilic - University of Tsukuba (Japan) [presenting]
Abstract: A study of the fuzzy clustering-based Principal Component Analysis (fuzzy clustering-based PCA) is presented which is capable of treating high-dimension, low-sample size data (HDLSS data) with high performance compared with ordinary PCA. In general, HDLSS data has difficulty analyzing by using conventional data reduction methods such as an ordinary PCA due to the inconsistency of eigenvalues of the sample covariance matrix with respect to variables, although one of the purposes of this analysis is the reduction of the number of dimensions (variables). In addition, in fuzzy clustering, the status of the clustering result shows not only whether the object belongs to clusters but also how much the object belongs to clusters. This can consider the practical situation of data. Therefore, the fuzzy clustering-based PCA can tackle the problem of ordinary PCA for the HDLSS data by including the scale of the result of fuzzy clustering. In particular, an application of this fuzzy clustering-based PCA is shown for the discrimination of individual subjects observed by sensors worn on the body during several activities. The analysis of this data is useful for healthcare, considering the individuality of the history of activities.
