model fit, overfit, and evaluation

Question 1

overfitting

Answer 1

the tendency to tailor models to the training data

Question 2

hold out validation and holdout accuracy

Answer 2

holdout accuracy is an estimate of generalization accuracy.
holdout data = test data

Question 3

why is overfitting bad

Answer 3

more complex -> spurious correlations (not representative)
product incorrect generalizations

Question 4

avoid overfitting for tree induction

Answer 4

1. stop growing the tree before it gets too complex
2. prune back a tree