6. Shrinkage methods Flashcards
1
Q
Purpose of shrinkage methods
A
Shrinking coefficients improves the fit because it can reduce variance.
2
Q
Ridge Regression
A
minimize SSR with L2 penalty term
( standardization necessary )
3
Q
LASSO
A
minimize RSS with L1 penalty term
4
Q
Test robustness (stability selection)
A
Counting proportion
( Sub-sampling and counting the proportion of times a variable is selected )
5
Q
Selecting tuning parameter
A
Minimal Cross-validation error rate
6
Q
Ridge vs. LASSO vs. Linear
A
In the case of multicollinearity or overfitting, Ridge or LASSO should be used. Both sacrifice predictive accuracy for simplicity, by putting a cost on model complexity.
