Comparing Logit and Hinge Surrogate Loss Functions in Outcome Weighted Learning

Comparing Logit and Hinge Surrogate Loss Functions in Outcome Weighted Learning

Bibliographic Details
Main Author: Eisner, Mariah Claire
Language:English
Published: The Ohio State University / OhioLINK 2020
Subjects:
Biostatistics
outcome weighted learning
surrogate loss function
hinge loss
logit loss
Online Access:http://rave.ohiolink.edu/etdc/view?acc_num=osu1585657996755039
id ndltd-OhioLink-oai-etd.ohiolink.edu-osu1585657996755039
record_format oai_dc
spelling ndltd-OhioLink-oai-etd.ohiolink.edu-osu15856579967550392021-08-03T07:14:07Z Comparing Logit and Hinge Surrogate Loss Functions in Outcome Weighted Learning Eisner, Mariah Claire Biostatistics outcome weighted learning surrogate loss function hinge loss logit loss Outcome weighted learning is a weighted classification-based approach for finding the optimal individualized treatment regime to prolong survival when subject characteristics impact response to different treatment options. Previous research on this method utilizes the hinge loss from machine learning to perform classification. However, there are other loss functions for binary classification that could be leveraged, such as the logit loss from logistic regression. This study compares the performance of outcome weighted learning models via simulations with different surrogate loss functions to determine whether the logit loss is a reasonable alternative to the hinge loss. Data are right censored with two possible treatments and decision functions are assumed to be linear. Simulations are conducted under three forms of the true decision function, using a correctly specified model with two covariates and an incorrectly specified model with an extra nuisance covariate. Logit loss and hinge loss outcome weighted learning models are applied to data from a randomized trial on aortic stenosis. Results indicate that the logit loss offers comparable performance to the hinge loss and therefore can be used as an alternative to the latter, though the performance of both outcome weighted learning models may suffer from instability when they are misspecified. 2020-10-01 English text The Ohio State University / OhioLINK http://rave.ohiolink.edu/etdc/view?acc_num=osu1585657996755039 http://rave.ohiolink.edu/etdc/view?acc_num=osu1585657996755039 unrestricted This thesis or dissertation is protected by copyright: all rights reserved. It may not be copied or redistributed beyond the terms of applicable copyright laws.
collection NDLTD
language English
sources NDLTD
topic Biostatistics
outcome weighted learning
surrogate loss function
hinge loss
logit loss
spellingShingle Biostatistics
outcome weighted learning
surrogate loss function
hinge loss
logit loss
Eisner, Mariah Claire
Comparing Logit and Hinge Surrogate Loss Functions in Outcome Weighted Learning
author Eisner, Mariah Claire
author_facet Eisner, Mariah Claire
author_sort Eisner, Mariah Claire
title Comparing Logit and Hinge Surrogate Loss Functions in Outcome Weighted Learning
title_short Comparing Logit and Hinge Surrogate Loss Functions in Outcome Weighted Learning
title_full Comparing Logit and Hinge Surrogate Loss Functions in Outcome Weighted Learning
title_fullStr Comparing Logit and Hinge Surrogate Loss Functions in Outcome Weighted Learning
title_full_unstemmed Comparing Logit and Hinge Surrogate Loss Functions in Outcome Weighted Learning
title_sort comparing logit and hinge surrogate loss functions in outcome weighted learning
publisher The Ohio State University / OhioLINK
publishDate 2020
url http://rave.ohiolink.edu/etdc/view?acc_num=osu1585657996755039
work_keys_str_mv AT eisnermariahclaire comparinglogitandhingesurrogatelossfunctionsinoutcomeweightedlearning
_version_ 1719456978284052480

Similar Items