A técnica LASSO e suas potencialidades na seleção de variáveis para modelos lineares

• Autor(a) principal: Silva, Carina Brunehilde Pinto da
• Data de Publicação: 2018
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Ceará (UFC)
• dARK ID: ark:/83112/001300000bvjh
• Texto Completo: http://www.repositorio.ufc.br/handle/riufc/41611
• topic: LASSO; Modelos de regressão; Seleção de variáveis; Regressão L1; Penalização; LASSO; Selection models; Selection of variables; Regression L1; Penalty

This research seeks to make a bibliographic survey of LASSO technique, Least Absolute Selection and Shrinkage Operator, as well as publicize the production of Brazilian studies in the area. This technique is important as an alternative for the adjustment of models with specific characteristics that make it difficult to use the estimation by least squares method, for which it uses penalties called L1. Throughout the work we use algebraic and geometric approaches to facilitate the understanding of the estimation process of both the coefficients of the models and the tuning parameter. We devote most of the work to discussing some of the disadvantages of LASSO, as well as the alternatives and generalizations that are emerging to circumvent them. We conclude the discussions by presenting the glmnet package, implemented in R software for adjustments using lasso, through two examples, with which they were possible to discuss the main functions of the package.