Wednesday, March 5, 2014

Is one model "SIGNIFICANTLY" better than another? use ANOVA

If model A has lower MSE (mean squared error) than model B does, A is better, right?

not necessarily, since you have a sample, not population, as your training set, which means your models don't have a complete picture.

Now what do we do?
Use ANOVA (analysis of variance).

"When given a single argument it produces a table which tests whether the model terms are significant.

When given a sequence of objects, anova tests the models against one another in the order specified."

No comments:

Post a Comment