On question 13 (cross-validation).
A few clarifications. First, this is not an analysis of data but it is meant as a reasoning exercise.
The idea is you should critique the issues which would arise in a selection/assessment approach. I also
have rephrased a little for clarity
-
Screen the predictors: perform separate analysis for each predictor to determine
whether it is correlated with the class label. Select a subset of “good” predictors
that show fairly strong correlation with the class labels (e.g. correlation > 0.8)
-
Using just this subset of predictors, choose the optimum tuning parameter of the classifier using cross validation
-
Estimate the prediction error of the model with the optimum tuning parameter using cross validation