Accession Number:

ADA606605

Title:

Report for Contract W911NF-09-1-0205 (University of Wisconsin - Madison)

Descriptive Note:

Final rept. 29 Apr 2009-31 Dec 2013

Corporate Author:

WISCONSIN UNIV MADISON

Personal Author(s):

Report Date:

2014-01-18

Pagination or Media Count:

11.0

Abstract:

Classification and regression tree methodology is an important and essential tool in statistics and machine learning. This research accomplished several improvements and advancements in the area and implemented them in the GUIDE computer software. The major contributions are i a new technique to deal with missing data values that allows all the information, including whether or not an observation is missing, to be used for tree construction and prediction, ii a new method of scoring the importance of variables that can be used to objectively reduce the number of variables for prediction modeling, iii a new approach to building regression models for data with multidimensional or longitudinal response variables that does not require any model assumptions, and iv several new techniques for identifying subgroups of the data for enhanced differential treatment effects.

Subject Categories:

  • Statistics and Probability

Distribution Statement:

APPROVED FOR PUBLIC RELEASE