Lecture 7 Imputation

Question 1

Model- driven imputation

Answer 1

Train regression model for jssung values
Retrain it after filling in
Flexible

Question 2

Knn imputation

Answer 2

Efficient implementation would find nearest neighbors only once
Naive — requires the first 2 always be presen t
Tricky if there is no feature that is always non- misusing

Question 3

Fan yum pure

Answer 3

No fit transform paradigm
Mice is iterative and works well
Nice might be best

Question 4

Feature selection

Question 5

Unsupervised

Answer 5

May discard important information
Variance- based: 0 variance or few unique value

Covariance based: remove correlated features
PCA: remove linear Subspaces

Question 6

Univariate selection

Answer 6

Examine each feature individually to determine the strength of the relationship of feature with two response variable —> provide a score for each feature

F score, chi2

Question 7

Mutual information

Answer 7

Univariate— doesn’t assume a linear model

Question 8

Multivariate — model based

Answer 8

Get best fit for a model
Exhaustive search — infeasible
Linear model assume linear relation

Question 9

Interactive model based

Answer 9

Fit model find least important feature,remove, iterate —- recursive feature elimination

Or start with with single feature, find most important, add