MED-ENS: Upgrading machine learning ensembles to address medical challenges

Ensemble methods are widely used machine learning prediction algorithms, with an excellent predictive performance, combined with feasible computation times. They are used in many application areas, including medicine. This domain is confronted with several challenges that prohibit an optimal application of ensemble methods. These challenges include missing class labels (originating from patients that are difficult to diagnose or from the time consuming labeling process), outcomes of a different data type than classically addressed by machine learning methods, the need for interpretable models, etc. In this project, we will address these challenges by upgrading ensemble methods. The models will be validated on datasets related to a complication of critical illness witnessed at an intensive care unit.