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

Figure 9

(a) Performance of logistic regression on real data. The graphic depicts 95% confidence intervals for the respective AUC value calculated and on the basis of [37]. All correction approaches perform similarly and significantly better than no correction (test by [37], )

(b) Performance of random forest on real data. The graphic depicts 95% confidence intervals for the respective AUC value calculated and on the basis of [37]. Only one correction approach, our novel parametric IP bagging, performs significantly better than no correction (test by [37], )

(c) Performance of logistic regression with all two-way interaction terms on real data. The graphic depicts 95% confidence intervals for the respective AUC value calculated and on the basis of [37]. All correction approaches perform significantly better than no correction (test by [37], )

(d) Performance of naive Bayes on real data. The graphic depicts 95% confidence intervals for the respective AUC value calculated and on the basis of [37]. All correction approaches perform significantly better than no correction (test by [37], )