Título: Tratamento de Características Ausentes via Subespaços Aleatórios e Imputação
Autores: Ribeiro, Mirlem Rodrigues; Santos, Eulanda Miranda dos
Resumo: Databases with missing features are very frequent in processing and pattern recognition real applications, as well as in other fields such as data mining. The most frequent solution employed in the literature to deal with missing features is based on substituting missing values with meaningful estimates. This is the so-called imputation of missing values. In this paper, we propose to combine imputation methods with classifier ensembles generated by random subspace in order to reduce data corruption caused by imputation. We present experimental results obtained using different databases. These databases range from relatively high-dimensional feature spaces to small feature spaces. Our results show that classifier ensembles generated by random subspace help to reduce data corruption and lead to better performance.
Palavras-chave: Classification methods; ensemble of classifiers; random subspace; missing features; imputation methods
Código DOI: 10.21528/CBIC2013-330
Artigo em pdf: bricsccicbic2013_submission_330.pdf
Arquivo BibTex: bricsccicbic2013_submission_330.bib