A0397
Title: Quadratic classifiers for high-dimensional noisy data
Authors: Aki Ishii - Tokyo University of Science (Japan) [presenting]
Kazuyoshi Yata - University of Tsukuba (Japan)
Makoto Aoshima - University of Tsukuba (Japan)
Abstract: One of the features of modern data is that the data dimension is extremely high. However, the sample size is relatively low. Such data is called HDLSS data. In HDLSS situations, new theories and methodologies are required to develop statistical inferences. High-dimensional classification is considered for noisy data such as genome data. It is noted that eigenvalues of high-dimensional noisy data grow very rapidly depending on the dimension. These eigenvalues obscure differences between populations. Two types of high-dimensional eigenvalue models exist the strongly spiked eigenvalue (SSE) model and the non-SSE (NSSE) model. High-dimensional classification under the SSE model is considered. New classifiers are given by using a data transformation technique. It is shown that our classifiers have preferable properties in theory. Finally, the classifiers are applied to real genome data sets.
