Seleção de modelos de regressão linear em bases de alta dimensão

Abstract

In many cases, it is necessary to deal with databases with a large number of variables, bases called Big Data. In some situations it is necessary to reduce the number of these variables, either to have a model with only the significant variables or to solve problems such as multicollinearity. With this, several methods have emerged to solve the above cases. This work proposed to study the different selection methods in multiple linear regression models, such as methods of selecting subsets of variables, methods of regularization and methods of dimensionality reduction. Cross-validation procedures were used in the regularization processes for selecting the models. Real data applications were performed using the R Core Team software (2018).