Analysis of copula based variable clustering techniques and application of mortality estimation

This paper aims at developing different mortality estimation models in MIMIC-III dataset. One of the aims of the study is to bring an efficient technical proposal to determine the dependency structures between the variables. The study is conducted with 38,015 adult intensive care patients in the MIMIC-III database. The dependency structure between the variables is determined and divided into clusters with CoClust and tail dependency. With logistic regression analysis applied through clusters, the number of significant and appropriate models for death variable within 24 hours was four while there were five for death variable in the hospital. When the obtained models were analysed with error matrix, cross validity criterion and ROC curve, three valid models were obtained for the death variable within 24 hours and two for the death variable in the hospital.