The use of data mining techniques to predict mortality and length of stay in an ICU

Abstract

Data mining is commonly used in the healthcare industry and managing Intensive Care Unit (ICU) is no exception. This study aims to examine how data mining techniques can be employed to predict mortality and length of stay in an ICU and to evaluate various classification techniques. Real-life healthcare datasets, like MIMIC 2, incorporate an unbalanced distribution of sample sizes, which means that it is difficult to employ them to assess classification. This paper presents an analysis of a mortality prediction algorithm to evaluate the extent to which this algorithm can predict mortality rate. The model aims to facilitate the process by which medical practitioners provide customized and optimized care in the ICU.