(A) study on improvement of high-dimensional function-based additive model

Abstract

In this paper, we introduce a Sparse Partially Linear Additive Models(SPLAM) to analyze a high-dimensional data. This is a model useful for variable selection and for checking if linearity of an independent variable is appropriate for explaining the dependent variable in an additive Model. We also introduce a method of bias reduction in parameter estimation by the Lasso. We introduce a new method of applying this method to the SPLAM to reduce the bias of parameter estimation in the SPLAM. To compare these two models, we use 3 loss functions for each data set. It is shown through experiments that the new method results in a decrease in error rate over the SPLAM.