model selection
When building a predictive model, it’s tempting to use an automated model selection method. For example, if you have 8 candidate predictors, there are 28 = 256 possible models to choose from. And that’s only if you consider main effects. If you include all 28 possible two-way interactions, the number of possible models grows into the millions. Clearly an exhaustive search of the “best” model becomes impractical and we’re motivated to somehow automate the process.