Correcting Classifiers for Sample Selection Bias in Two-Phase Case-Control Studies

Algorithm 2

Parametric inverse-probability bagging.

Input: Observed sample of size , IP weights

Output: Unbiased prediction for new unbiased data

(1) for    to    do

for    to    do

(a) Estimate parameters of distribution

(b) Draw parametric bootstrap sample from of size

(c) Rebuild stratum as , where “” denotes -fold concatenation

end

(a) Combine strata to sample:

with

(b) Fit classifier

end

(2) Output the ensemble of learners

(3) Aggregate predictions on new data set by averaging: