Shrinkage-based classification methods for analyzing high-dimensional data

With the advent of technologies, learning high-throughput complex models is becoming critical in many disciplines. It is known that such high-dimensional data pose challenges to traditional statistics methods and motivate more accurate and powerful statistical methods. In this proposal, the investigators propose to use shrinkage techniques that aim to improve the performance of class prediction in high-dimensional data analysis. Specifically, in this research plan, the following two related projects are going to be investigated: (i) Bias Correction for Shrinkage-based Discriminant Rules; and (ii) Optimal Nearest Shrunken Centroids Method.

The proposed research in this proposal is an essential step in the development of shrinkage theory and methods in high-dimensional data analysis. The preliminary studies by the investigators indicate that the proposed methods are both statistically efficient and computationally inexpensive. Finally, we reiterate that the proposed methods have extensive applications and can be applied to data sets from various areas including statistical genetics, epidemiology, ecology, brain imaging, and engineering sciences.