and excluded in the "best" model. For each method/model you

fit, report the three most important predictive variables (they may not

be the same across methods).

Note: If you use neural networks with the nnet package, the olden function

from the "NeuralNetTools" packages can be used to assess relative importance

of the variables in the model. You can get variable importance from a decision

tree with mod$variable.importance for the decision tree or from the summary

function.

For the each method/model fit, report the training data area under the

curve (AUC) and the test data AUC.

Answer the following questions

For each method/model fit, is the training AUC better or worse than the

test AUC?

In comparing the predictive ability of the two models with each other, is

better to compare the AUC from the training data or from the test data?

Why?

In general, is the test AUC expected to be larger or smaller than the test

AUC? Why? Is this always the case?

Which of the two methods/models fits/performs better?

1