Vital Prognosis of Patients in Intensive Care Units Using an Ensemble of Bayesian Classifiers

An Ensemble of Bayesian Classifiers (EBC) is constructed to perform vital prognosis of patients in the Intensive Care Units (ICU). The data are scarce and unbalanced, so that the size of the minority class (critically ill patients who die) is very small, and this fact prevents the use of accuracy as a measure of performance in classification; instead we use the Area Under the Precision-Recall curve (AUPR). To address the classification in this setting, we propose the use of an ensemble constructed from five base Bayesian classifiers with the weighted majority vote rule, where the weights are defined from AUPR. We compare this EBC model with the base Bayesian classifiers used to build it, as well as with the ensemble obtained using the mere majority vote criterion, and with some state-of-the-art machine learning supervised classifiers. Our results show that the EBC model outperforms most of the competing classifiers, being only slightly surpassed by Random Forest.