• predictions
• von
• vermont
• oxford
• cross-validation
• models
• network
• preprocessing
• roc

Mortality predictions of low and very low birth-weight infants using logistic regression models are widely used in risk adjustment procedures for comparing different NICUs (Neonatal Intensive Care Units).
We tackled this problem from a machine learning point of view by training state-of-the-art supervised models for the task. Furthermore, we used unsupervised techniques to provide clinicians with new insights on the matter that could ultimately lead to new improvements.