Robust mislabel logistic regression without modeling mislabel probabilities

Robust mislabel logistic regression without modeling mislabel probabilities IntroductionLogistic regression is one of the most widely used statistical methods for linear discriminant analysis. Let Y0 be a binary response with {0,1} values, and X be the p‐dimensional random vector of explanatory variables. Logistic regression assumes P(Y0=1|X=x) to satisfy the conditional label probability model π(x;β)=exp(β⊤x)1+exp(β⊤x), where β=(β1,…,βp)⊤, and let β0 denote the true value of β in model . The MLE is known to be the most efficient estimator for β0 when data are truly generated from model . However, in some situations we can only observe a contaminated label Y instead of the true status Y0. That is, Y is flipped from Y0 according to the mislabel probabilities η0(x)=P(Y=1|Y0=0,X=x) andη1(x)=P(Y=0|Y0=1,X=x). The success probability of Y no longer follows model , but instead has the form P(Y=1|X=x)=η0(x){1−π(x;β)}+{1−η1(x)}π(x;β). Fitting label contaminated data {(Yi,Xi)}i=1n to the uncontaminated model will produce a biased estimate of β0. To overcome the problem of mislabeling, some robustified logistic regression methods are developed based on with different modelings for ηj(x)'s. Copas () considered equal and constant mislabel probabilities, η0(x)=η1(x)=η, which we call the constant‐mislabel logistic regression. For any given η, the estimating equation of β is 1n∑i=1nwη,i(β){Yi−πη(Xi;β)}Xi=0 with πη(x;β)=η{1−π(x;β)}+(1−η)π(x;β) and the weight function wη,i(β)=1−2η{1−η+ηexp(−β⊤Xi)}{1−η+ηexp(β⊤Xi)}. Another example http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Biometrics Wiley

Robust mislabel logistic regression without modeling mislabel probabilities

Loading next page...
 
/lp/wiley/robust-mislabel-logistic-regression-without-modeling-mislabel-Btm2V5Q0RR
Publisher
Wiley
Copyright
© 2018, The International Biometric Society
ISSN
0006-341X
eISSN
1541-0420
D.O.I.
10.1111/biom.12726
Publisher site
See Article on Publisher Site

Abstract

IntroductionLogistic regression is one of the most widely used statistical methods for linear discriminant analysis. Let Y0 be a binary response with {0,1} values, and X be the p‐dimensional random vector of explanatory variables. Logistic regression assumes P(Y0=1|X=x) to satisfy the conditional label probability model π(x;β)=exp(β⊤x)1+exp(β⊤x), where β=(β1,…,βp)⊤, and let β0 denote the true value of β in model . The MLE is known to be the most efficient estimator for β0 when data are truly generated from model . However, in some situations we can only observe a contaminated label Y instead of the true status Y0. That is, Y is flipped from Y0 according to the mislabel probabilities η0(x)=P(Y=1|Y0=0,X=x) andη1(x)=P(Y=0|Y0=1,X=x). The success probability of Y no longer follows model , but instead has the form P(Y=1|X=x)=η0(x){1−π(x;β)}+{1−η1(x)}π(x;β). Fitting label contaminated data {(Yi,Xi)}i=1n to the uncontaminated model will produce a biased estimate of β0. To overcome the problem of mislabeling, some robustified logistic regression methods are developed based on with different modelings for ηj(x)'s. Copas () considered equal and constant mislabel probabilities, η0(x)=η1(x)=η, which we call the constant‐mislabel logistic regression. For any given η, the estimating equation of β is 1n∑i=1nwη,i(β){Yi−πη(Xi;β)}Xi=0 with πη(x;β)=η{1−π(x;β)}+(1−η)π(x;β) and the weight function wη,i(β)=1−2η{1−η+ηexp(−β⊤Xi)}{1−η+ηexp(β⊤Xi)}. Another example

Journal

BiometricsWiley

Published: Jan 1, 2018

Keywords: ; ; ; ;

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off