Uma variante do método de regressão Forward Stagewise para dados incompletos

Abstract

The Forward Stagewise regression is a popular regression method to achieve sparse models, that is, a model where some regression coefficients are null. However, in its formulation only fully observable databases are considered. As in many cases, available data do not always follow such hypothesis, and it is necessary that they pass through some kind of treatment in a pre-processing step or a method produced for the purpose of a particular characteristic must be used. In this work we extend the Forward Stagewise regression to treat cases where missing data is present at the input of the method. Such extension is obtained by including in the method the procedure for treating the records which have missing data so that the uncertainty of those data is considered. Beyond in this work, we performed a set of tests on the real world’s databases in which we obtained positive results.