Bayesian nutritional model for morbidity prognosis in newborns

Abstract

This research aimed to formulate a Bayesian model based on the Naive Bayes algorithm, to predict morbidity in neonates in a case study of pregnant mothers in Metropolitan Lima. The study uses mathematical algorithms for the exploitation of information in prevention of possible health-related problems. 13 predictive nutritional variables proposed by Krauss were raised. The model consists first of all, in the collection of the nutritional information in a controlled way of the pregnant women involved, then, the information is analyzed to determine the relationship of the most influential variables for the model, then the Bayesian model of acyclic characteristic was constructed and directed composed of nodes and edges, because the variables directly affected to the morbidity of the neonate are known and finally the model affected by the statistical results of the nutritional variables is validated, as part of the process of formulating the model and by experts judgment in the topic. The results conclude that the predictive variables that directly influence are: breads, sugars, oils, fats and salt; and conversely: fruits, water, vegetables and vegetables; the model also predicts the morbidity of the newborn with a probability of 92% and an error of 8.0%.

Keywords

• newborn morbidity
• bayesian networks
• morbidity prognosis