Feature selection is an important step to reduce data dimensionalityand hopefully improve classification accuracy. In this paper we implemented a"visual" wrapper+Sequential Forward Search (SFS) procedure usign R/Netlogoon two datasets, Breast Cancer Diagnosis from UCI repository (WBCD) andIBOVESPA daily data from 2001 to 2015, retrieved from YAHOO Finance withtwenty eight technical indicators added to the data. In the IBOVESPA datasetwe also used a wrapper+Sequential Backward Search (SBS) procedure and ahybrid method, combining a filter with SBS for five different filters. All of themused a Support Vector Machine (SVM) with a (Gaussian) Radial Basis Functionkernel (RBF) and grid-search optimization.
http://www.lbd.dcc.ufmg.br/colecoes/eniac/2016/001.pdfCaso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web