OPTIMIZATION CHALLENGE: AREA UNDER RECEIVER OPERATING CHARACTERISTIC AS A LOSS FUNCTION IN GENERALIZED LINEAR MODELS

Result code in IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26220%2F10%3APU88680" target="_blank" >RIV/00216305:26220/10:PU88680 - isvavai.cz</a>
Result on the web
—
DOI - Digital Object Identifier
—

Result language
angličtina
Original language name
OPTIMIZATION CHALLENGE: AREA UNDER RECEIVER OPERATING CHARACTERISTIC AS A LOSS FUNCTION IN GENERALIZED LINEAR MODELS
Original language description
The specific group of models are binary regression classifiers, e.g. the logit model. These models are built based on the quantitative loss functions like maximum likelihood estimation but their performance is obviously expressed by qualitative measure like accuracy or area under the receiver operating characteristic. In this paper we demonstrate the informational difference between quantitative a qualitative measures and suggest an alternative approach to setup of linear models based on the qualitativemeasure and genetic algorithms. We achieved statistically significant improvement of linear model setup by use of genetic algorithm, which overperforms equivalent logit model.
Czech name
—
Czech description
—

Project
<a href="/en/project/GA102%2F09%2F1897" target="_blank" >GA102/09/1897: Safety of Car Transportation</a><br>
Continuities
Z - Vyzkumny zamer (s odkazem do CEZ)

Publication year
2010
Confidentiality
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Similar results(10)