Una medida de influencia en la clasificación a través de la teoría de juegos cooperativos
Loading...
Identifiers
Publication date
Authors
Advisors
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
[ES] Partiendo de un conjunto de datos que han sido clasificados en dos clases, queremos
saber cuáles han sido los factores más importantes para determinar la clase que les ha sido
asignada. Para ello se usará una aproximación axiomática con la finalidad de definir una
medida de influencia única. Esta medida será una función que, a partir de un conjunto de
datos clasificados, devolverá un valor para cada característica correspondiente a la influencia que ha tenido a la hora de determinar la clasificación. Primero se comenzará con una
breve introducción a los juegos cooperativos con utilidad transferible, ya que axiomatizar
una medida de influencia y una solución de teoría de juegos tienen características comunes,
y varios modelos de clasificación. Después de definir la medida de influencia, se muestra
que tiene una forma intuitiva cuando el clasificador es lineal. Y por último, aplicamos la
medida en un conjunto de datos relacionados con pacientes afectados por el COVID-19
dependiendo de si han tenido algún incidente o no.
[EN] Starting from a collection of data that have been classified into two classes, we want to know which factors have been the most important in determining the classification outcome. For this purpose, an axiomatic approach will be used in order to define a unique measure of influence. This measure will be a function that, given a set of classified data, outputs a value for each feature corresponding to its influence in determining the classification outcome. We will first begin with a brief introduction to cooperative games with transferable utility, since axiomatizing an influence measure and a game theory solution have common aspects, and several classification models. After defining the influence measure, we show that it has an intuitive form when the classifier is linear. Finally, we apply the measure on a dataset related to patients affected by COVID-19 depending on whether they have had any incident or not.
[EN] Starting from a collection of data that have been classified into two classes, we want to know which factors have been the most important in determining the classification outcome. For this purpose, an axiomatic approach will be used in order to define a unique measure of influence. This measure will be a function that, given a set of classified data, outputs a value for each feature corresponding to its influence in determining the classification outcome. We will first begin with a brief introduction to cooperative games with transferable utility, since axiomatizing an influence measure and a game theory solution have common aspects, and several classification models. After defining the influence measure, we show that it has an intuitive form when the classifier is linear. Finally, we apply the measure on a dataset related to patients affected by COVID-19 depending on whether they have had any incident or not.
Description
Traballo Fin de Grao en Matemáticas. Curso 2020-2021
Keywords
Bibliographic citation
Relation
Has part
Has version
Is based on
Is part of
Is referenced by
Is version of
Requires
Sponsors
Rights
Atribución-NoComercial-CompartirIgual 4.0 Internacional



