Date of Award

2013

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Mathematics and Statistics

First Advisor

Adele Cutler

Abstract

Random Forests is a useful ensemble approach that provides accurate predictions for classification, regression and many different machine learning problems. Classification has been a very useful and popular application for Random Forests. However, it is preferable to have the probability of a membership rather than the simple knowledge that one belongs to whichever group. Votes and the regression method are current probability estimation methods that have been developed in Random Forests. In this thesis, we introduce two new methods, proximity weighting and the out-of-bag method, trying to improve the current methods. Several different simulations are designed to evaluate the new methods and compare them with the old ones. Finally, we use real data sets from UCI machine learning repository to further evaluate and compare those methods.

Share

COinS