A comparison of mortality prediction models for neonatal intensive care patients using artificial neural networks and case-based reasoning.

摘要

This thesis attempts to find the minimum variable set that gives the best neonatal mortality prediction using day 3 data from the neonatal intensive care unit (NICU). A hybrid system is used consisting of an artificial neural network and the kNN case-based reasoner, which is verified as a valid imputation tool for missing values.; Two minimum variable sets were derived by reducing variables from linear and nonlinear network structures. The linear model obtained a higher average sensitivity and both day 3 models obtained higher areas under the ROC when compared with other models.