A Comparative Analysis of the Neonatal Prognosis Problem Using Artificial Neural Networks, Statistical Techniques and Certainty Management Techniques

One of the most popular methods for assessing fetal well-being is the nonstress test (NST). The expert system NST-EXPERT, performs a diagnosis of the nonstress test and formulates therapeutic plans, while taking into account different aspects of the maternal-fetal context, and incorporating these analysis into a model for predicting fetal outcome. The prognosis module actually implemented in the NST-EXPERT uses a mathematical model (based on the certainty factors of Shortliffe and Buchanan) that combines the NST diagnosis with the risk factors present in the maternal-fetal context to predict the fetal outcome. In this work we describe another different approaches followed based on artificial neural networks (ANN), statistical techniques (Bayes, logistic regression) and certainty management methods (Dempster-Shafer). The validation of the different models is performed through a formal methodology of validation using the tool SHIVA.