A0618
Title: Feature selection for functional predictors
Authors: Dennis Cox - Rice University (United States) [presenting]
Abstract: In prediction problems with functional predictors the individual values of the functions typically have little predictive value. The problem of selecting features depends very much on the particular problem at hand. Several data sets with spectroscopic measurements of patient tissue are considered. The objective is to predict whether or not there is precancerous disease at the measurement site. A variety of methods for extracting potentially useful features are considered in the context of various machine learning algorithms that are employed for the prediction. Some simple, classical methods perform quite well for these applications.
