Description
We combine bootstrap aggregation and imputation to capture the uncertainty in the imputed missing values and reduce the variability in the final learned model. We evaluate this method in a healthcare dataset comprised of patients who have undergone a stress echocardiography test for the purpose of predicting 5-yr mortality. We find that ensemble imputation improves performance compared to conventional single imputation.
Learning Objective: Formulate an imputation approach which accounts for uncertainty of imputed values and evaluate the impact of the uncertainty.
Authors:
David Chen (Presenter)
Mayo Clinic
Christopher Scott, Mayo Clinic
Christina Luong, Mayo Clinic
Itzhak Attia, Mayo Clinic
Che Ngufor, Mayo Clinic
Adelaide Arruda-Olson, Mayo Clinic
Patricia Pellikka, Mayo Clinic
Presentation Materials: