Fig. 3.
L1 regularized logistic regression, random forest, and gradient boosted tree model were trained using general and labs features, for a single random split of the data. The precision and recall are computed from the test set. For comparison, the curves for the HOSPITAL (low hemoglobin at discharge, discharge from an oncology service, low sodium at discharge, [International Classification of Disease, ninth revision, clinical modification coded] procedure during hospital stay or not, index admission type is elective or not, number of admissions in previous year, and length of stay) and LACE (length of stay, acuity, comorbidity, emergency department use) scores are also shown.