Abstract
Predicting mortality rate for the patients in Intensive Care Unit (ICU) is an active topic in medical domain for decades. The main goal of mortality prediction is to achieve satisfied discrimination and calibration. However, the particular features of the patient records such as high-dimension, irregular, and imbalance nature of ICU data makes prediction challenging. Data mining is gaining an ever-increasing popularity in predicting mortality of ICU patients recently, a comprehensive literature review of the subject has yet to be carried out. This study presented a review of and classification scheme for the past research as well as latest progress and their limitations on application of data mining techniques for predicting ICU mortality. Based on limitations, a hybrid framework combined with intrinsic property of ICU data to improve prediction performance is proposed for future research.