Impact of precision of Bayesian network parameters on accuracy of medical diagnostic systems

Abstract

Objective One of the hardest technical tasks in employing Bayesian network models in practice is obtaining their numerical parameters. In the light of this difficulty, a pressing question, one that has immediate implications on the knowledge engineering effort, is whether precision of these parameters is important. In this paper, we address experimentally the question whether medical diagnostic systems based on Bayesian networks are sensitive to precision of their parameters. Methods and Materials The test networks include Hepar II, a sizeable Bayesian network model for diagnosis of liver disorders and six other medical diagnostic networks constructed from medical data sets available through the Irvine Machine Learning Repository. Assuming that the original model parameters are perfectly accurate, we lower systematically their precision by rounding them to progressively courser scales and check the impact of this rounding on the models' accuracy. Results Our main result, consistent across all tested networks, is that imprecision in numerical parameters has minimal impact on the diagnostic accuracy of models, as long as we avoid zeroes among parameters. Conclusion The experiments' results provide evidence that as long as we avoid zeroes among model parameters, diagnostic accuracy of Bayesian network models does not suffer from decreased precision of their parameters.

Modeling the viability of the free‐ranging cheetah population in Namibia: an object‐oriented Bayesian network approach

“…This information may be biased and based on intuition rather than real data. However, many parameters do not require great precision (Onisko and Druzdzel 2011) and expert opinions are ideally suited to estimate them. Nonetheless, it is critical for the authenticity and predictive accuracy of the BN inference to refine those parameters that are shown to have a more profound effect on the target node (Onisko and Druzdzel 2011).…”

Decision Support System for Hepatitis Disease Diagnosis using Bayesian Network

“…A Bayesian belief network is also known as acyclic graphical model. It is a probabilistic model that denotes a set of arbitrary variables and their conditional independencies through a directed acyclic graph [12]. A Bayesian Network could denote the probabilistic connections between disorders and symptoms.…”

Impact of Bayesian Network Model Structure on the Accuracy of Medical Diagnostic Systems

“…In our earlier study, focusing on the impact of precision of numerical parameters on the quality of Bayesian network results [5], we selected six medical data sets from the Irvine Machine Learning Repository: Acute inflammation [9], SPECT Heart, Cardiotocography, Hepatitis, Lymphography [10], and Primary Tumor [10]. We used the following two selection criteria: (1) the data set had to have at least one disorder variable and (2) it should not contain too many missing values and too many continuous variables.…”

“…There is a popular belief that it is the structure of Bayesian networks that is important and that they are insensitive to the overall noise and precision of their numerical probabilities. There is a body of empirical work showing that indeed the precision of numerical parameters is not important to the quality of results (e.g., [3,4,5,6]). To our knowledge, there has been no parallel work testing the importance of graphical structure of Bayesian networks.…”