7.4 Practical Considerations
Is it possible to identify all possible predictive interactions?
Is it possible to evaluate all possible interactions?
Should the interaction terms be created before or after the preprocessing part?
Only a fraction of all possible pairwise interactions contain relevant information.
With \(p\) predictors we have \((p)(p-1)/2\) pairwise interaction terms.
Here is the difference in interaction effects before and after preprocessing: