Date of Award

Fall 5-5-2017

Degree Type

Creative Project

Degree Name

Master of Science (MS)

Department

Mathematics and Statistics

First Advisor

Richard Cutler

Second Advisor

Chris Corcoran

Third Advisor

Luis Gordillo

Abstract

Survival analysis methods are a mainstay of the biomedical fields but are finding increasing use in other disciplines including finance and engineering. A widely used tool in survival analysis is the Cox proportional hazards regression model. For this model, all the predicted survivor curves have the same basic shape, which may not be a good approximation to reality. In contrast the Random Survival Forests does not make the proportional hazards assumption and has the flexibility to model survivor curves that are of quite different shapes for different groups of subjects. We applied both techniques to a number of publicly available datasets and compared the fit of the two techniques across the datasets using the concordance index and prediction error curves. In this process we identified 'types of data' in which Random Survival Forests may be expected to outperform the Cox model.