Problemas de regresión e clasificación usando SVM

Loading...
Thumbnail Image
Identifiers

Publication date

Advisors

Editors

Journal Title

Journal ISSN

Volume Title

Publisher

Metrics
Google Scholar
lacobus
Export

Research Projects

Organizational Units

Journal Issue

Abstract

[GL] O noso obxectivo é estudar un tipo de algoritmos de aprendizaxe coñecidos como support vector machines. Comezamos por introducir un caso sinxelo de clasificación binaria, o clasificador de marxe máxima, que empregamos para clasificar datos linealmente. Despois presentamos o concepto de función kernel, importante á hora de realizar clasificacións non lineais e introducimos as variables slack para mellorar a capacidade de predición. A continuación, estudamos dous métodos que nos permiten traballar con problemas de clasificación multiclase. No segundo capítulo xeneralizamos as support vector machines aos problemas de regresión mediante o uso de funcións de perda ε-insensitivas, que se basean en ignorar os erros se estes son o suficientemente pequenos. Ao final do traballo ilústrase a implementación en R destes algoritmos con dous exemplos reais.
[EN] Our objective will be studying a type of leaning algorithms known as support vectormachines. We start by introducing a simple example of binary classiffcation, the maximal margin classifier, which we use to separate data linearly. Then we present the kernel functions, which become really important at classifying data in a non-linear way, and slack variables are introduced to improve the prediction ability. Next, we study two methods that allow us to apply the algorithm to non-binary classification. In the second chapter, the concept of support vector machine is generalized to apply it to regression problems by defining an ε-insensitive loss function that ignores errors smaller than a certain constant. Finally, we illustrate how to apply these algorithms in R by using two real examples.

Description

Traballo Fin de Grao en Matemáticas. Curso 2018-2019

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