|
Abstract:
|
Irregular functional data, characterized by spatial heterogeneity and local features like peaks, are commonly encountered in biomedical research, such as spectrometric data. However, most of the current methods in functional data analysis are based on smoothing methods using global penalties and bandwidth. In this presentation, we proposed a novel feature extraction method which is particularly useful in detecting local features for classification of functional data. More specifically, it wisely selects a few "markers" over the spectrum and utilizes the discriminant coordinates on the subspace around these markers as the extracted features. We apply the method to two public domain data sets together with simulated data and discuss the understanding developed from these applications.
|
