Tree-Structured Methods for Prediction and Data Visualization

摘要

The aim of the research is to develop the GUIDE algorithm into a fast, powerful, and comprehensive procedure for tree-structured prediction and data visualization. During this reporting period, the regression component was enhanced by the addition of least squares regression through the origin, best simple analysis of covariance, all subsets regression, and least median of squares regression. An option to truncate the predicted values also was added. A preliminary classification tree component included kernel and nearest- neighbor node modeling. Numerous improvements were made to the algorithms for split and variable selection and importance scoring. GUIDE now supersedes the older CRUISE and QUEST algorithms. The GUIDE computer program had three major revisions and continues to be distributed for free over the Internet. Two PhDs were graduated and fifteen papers published or accepted for publication during this period.