Learning Bayesian Networks Using Heart Failure Data
Muzeyin Ahmed Berarti
Department of Statistics, Wolaita Sodo University, P.O.Box 138, Ethiopia
Ayele Taye Goshu *
Department of Statistics, School of Mathematical and Statistical Sciences, Hawassa University, P.O.Box 05, Ethiopia
*Author to whom correspondence should be addressed.
Abstract
Background: Several factors may affect heart failure status of patients. It is important to investigate whether or not the effects are direct. The purpose of this study was learning Bayesian networks that encode the joint probability distribution for a set of random variables.
Methods: The design was a retrospective cohort study. The target population for this study was
heart failure patients who were under follow- up at Asella referral teaching Hospital from February, 2009 to March, 2012. Bayesian Network is used in this paper to examine causal relationships between variables via Directed Acyclic Graph (DAG).
Results: Death of patients can be determined using HIV, hypertension, diabetes, anemia, renal inefficiency and sinus. Hypertension and sinus were found to have direct effects while TB had only indirect effect. Age did not have an effect.
Conclusion: Anemia, HIV, diabetes mellitus renal inefficiency and sinus directly affect the death of heart failure patient. Death is conditionally independent on TB and age, given all other variables.
Keywords: Bayesian network, parameter learning, structure learning, causal relationships