Applying statistical, uncertainty-based and connectionist approaches to the prediction of fetal outcome: a comparative study.

Abstract

A common situation in the field of medicine is the availability of a huge quantity of data and knowledge relevant to a problem which is nevertheless, to a greater or lesser degree, incomplete or imprecise. This kind of problem occurs, for example, with respect to the information available for the prognostic tasks required for pregnancy monitoring, where decision-making by physicians calls for the incorporation of predictive skills. There are available however, several knowledge discovery methods that can be applied to data resulting from the performance of one or several of the non-stress tests (NSTs) that are used to evaluate a pregnant patient's antenatal status. This paper presents, discusses and compares the results obtained as a consequence of the application of different prediction methods, namely the Bayes' model, discriminant analysis, artificial neural networks (ANNs) and the Shortliffe and Buchanan uncertainty-based model.