Fig. 1. Workflow for improving data balance in machine learning prognosis prediction using double prioritized (DP) bias correction.

Sample Enrichment prepares a number of new training datasets by incrementally enriching a specific demographic subgroup; Candidate Training is where each of the n + 1 datasets is used for training a candidate machine learning model; Model Selection identifies the optimal model; Prediction applies the selected model on new patient data. AUC-PR represents the area under the curve of the precision-recall curve.