Stringing: From High-Dimensional to Functional Data

• Activity Number: 106
• Type: Invited
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Hans-Georg Mueller*+
• Companies: University of California at Davis
• Keywords: Functional Data Analysis ; Correlated Predictors ; Large p, Small n ; Prediction ; Visualization ; Tree Rings

Abstract

Stringing takes advantage of high dimensionality of data vectors by representing such data as discretized and noisy observations that are generated from a smooth stochastic process. Assuming that data vectors result from scrambling the original ordering of the observations, Stringing proceeds by reordering the components of the high-dimensional vectors, thereby transforming the vectors into functional data. After this transformation, established techniques from Functional Data Analysis may be applied for subsequent statistical analysis, such as generalized functional linear regression or other functional regression models. This methodology provides an alternative to popular sparse selectors. Simulations show that Stringing leads to better prediction if key assumptions of sparse selection methods, in particular sparsity or near uncorrelatedness of the predictors, are violated. Data illustrations include gene expression and tree ring data. This talk is based on joint work with Kehui Chen, Kun Chen, Jane-Ling Wang and Ping-Shi Wu.