A simple R library for stepwise selection of the "best" linear model, depending on the selected criterion, such as adjusted R2, AIC, BIC, Mallows' Cp or p-values.
- forward selection, based on adjusted R2 (function
stepwise.fwd.adjR2) - backwards elimination, based on adjusted R2 (function
stepwise.bck.adjR2) - forward selection, based on Mallows' Cp (function
stepwise.fwd.mallowsCp) - backwards elimination, based on Mallows' Cp (function
stepwise.bck.mallowsCp) - forward selection, based on Aikake's Information Criterion (function
stepwise.fwd.aic) - backwards elimination, based on Aikake's Information Criterion (function
stepwise.bck.aic) - forward selection, based on Bayesian Information Criterion (function
stepwise.fwd.bic) - backwards elimination, based on Bayesian Information Criterion (function
stepwise.bck.bic) - forward selection, based on significance of coefficients' p-values (function
stepwise.fwd.pval) - backwards elimination, based on significance of coefficients' p-values (function
stepwise.bck.pval)
Just call the command source('stepwise.R') and then call any function
mentioned in the previous section. For more info, see "documentation"
comments in headers of those functions.
Alternatively see the file test.R that performs basic unit tests.
The functions do not (pre)process provided data frames in any way. In other words, the data frame must be preprocessed beforehand if required. This includes all desired transformations of variables, columns with factor variables must be provided as vectors of strings, additional columns (such as powers, interactions, etc.) must be added. It is also strongly recommended that undesired variables are removed from the data frame.
All files are are licensed under the Apache 2.0 license.
The author of the library is Jernej Kovačič.