ación de Métodos Variacionales para la Inferencia Estadística Bayesiana : inferencia variacional en modelos gráficos probabilísticos para el estudio de la relación entre enfermedades y síntomas.

Abstract

Medical diagnosis problems in primary care, where the relationship between diseases and symptoms must be considered, entails taking a significant amount of uncertainty and a large number of variables involved. In scenarios involving these two components, graphical models are an excellent alternative, since combining probability theory and graph theory results in strong multivariate statistical modelling. Therefore, in this work we will propose a probabilistic graphic model for the diseases: tuberculosis, pneumonia, allergy, bronchial asthma and the common flu, and their respective symptoms. Graphical models play an important role when exploring probability distributions of interrelated variables from a series of data; however, due to the difficulty in calculating a conditional probability distribution on the latent variables, taking into account the evidence, in our case the diseases given the symptoms we will introduce the variational inference that is part of the methods that brings the densities closer to through optimization, providing approximations for the marginal and conditional probabilities.