Variable Selection - Principal Approach

In principle there are two methods to select the variables for a multivariate model to be established:
 

• select the variables off-line: the variables are selected before the multivariate model is set up. This is simple but may not be efficient if the selection method does not fit the modeling method. Examples: selection by Fisher ratio, or selection according to correlations.

 

 


 
 

• select the variables including the modeling method: the variable selection is included in the whole modeling process. This ensures that the variables are selected in a way which gives the best results with the given modeling method. Examples of this approach are stepwise regression, or growing neural networks.