A comparison of inductive modeling techniques for pediatric decision making

Many medical decision problems have a number of characteristics that confound traditional approaches to data analysis, information processing and decision making. First, the data are limited and difficult to obtain. Second, there is a large number of variables that may or may not impact directly on the response variable of interest. Finally, historical information does not exist to guide either model identification of feature selection. This paper investigates the effectiveness of three feature-based inductive modeling techniques under such conditions. The specific problem we examine is cardiac output prediction in pediatric intensive care patients