A mixed quadratic programming model for a robust support vector machine.

Support Vector Machines are extensively used to solve classification problems in Pattern Recognition. They deal with small errors in the training data using the concept of soft margin, that allow for imperfect classification. However, if the training data have systematic errors or outliers such stra...

Descripción completa

Detalles Bibliográficos
Autores: Serna Diaz, Raquel, Leite, Raimundo Santos, Silva, Paulo José da Silva e
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2021
País:Brasil
Institución:Universidade Federal de Ouro Preto (UFOP)
Repositorio:Repositório Institucional da UFOP
Idioma:inglés
OAI Identifier:oai:repositorio.ufop.br:123456789/16117
Acceso en línea:http://www.repositorio.ufop.br/jspui/handle/123456789/16117
https://dx.doi.org/10.17268/sel.mat.2021.01.03
Access Level:acceso abierto
Palabra clave:Mixed integer quadratic programming
Outliers
Classification
Descripción
Sumario:Support Vector Machines are extensively used to solve classification problems in Pattern Recognition. They deal with small errors in the training data using the concept of soft margin, that allow for imperfect classification. However, if the training data have systematic errors or outliers such strategy is not robust resulting in bad generalization. In this paper we present a model for robust Support Vector Machine classification that can automatically ignore spurius data. We show then that the model can be solved using a high performance Mixed Integer Quadratic Programming solver and present preliminary numerical experiments using real world data that looks promissing.