MATLAB: Does MATLAB have any routines for doing Variable Regression Selection Procedures (VRSP)

"does(vrsp)?anydoingforhaveMATLABProceduresregressionroutinesselectionStatistics and Machine Learning Toolboxvariable

Does MATLAB have any routines for doing Variable Regression Selection Procedures (VRSP)?

Best Answer

In v2 of the Statistics Toolbox, there is a MATLAB file called REGSTATS that has many regression diagnostics. Also, there is a function called STEPWISE that will do forward and backward, as well as, ad-hoc stepwise regression.
As of Statistics Toolbox version 7.6 (R2011b), the following procedures are available for doing feature subset selection:
If you are doing cross sectional regression, you can view an introduction to feature selection with the Statistics Toolbox by executing the following command from the MATLAB command line:
>> web([docroot '/toolbox/stats/brj0qbu.html#brkgtb0'])
Stepwise model selection can be done using the function STEPWISEFIT in the Statistics Toolbox. This can be used for Stepwise and for Forward Selection. To do forward selection set the value of the parameter 'PREMOVE' to something just less than 1 (e.g. 0.999). This way features will not be removed from the set, only added. For more information about STEPWISEFIT, please execute the following command at the MATLAB command prompt:
>> web([docroot '/toolbox/stats/stepwisefit.html'])
There is also a GUI available for stepwise model selection which you can accessed through the function STEPWISE. This can be found at:
>> web([docroot '/toolbox/stats/sequentialfs.html'])
There is also a general function called SEQUENTIALFS which can be used to do forward and backward feature selection based on a user defined criterion. You may be able to use this to achieve functionality similar to the R-Squared Selection that you described. For more information, please execute the following command from the MATLAB command prompt:
>> web([docroot '/toolbox/stats/brj0qbu.html#brkgtb0'])
The LASSO function can also be used to reduce the number of features in a particular model. For more information about the LASSO command, please refer to the following link:
http://www.mathworks.com/help/releases/R2011b/toolbox/stats/lasso.html
Related Question