A suggestion for constructing a Bayesian network model with simple correlation and an appropriate regression analysis: A real medical diagnosis application

Abstract

The major task of medical science is to prevent or diagnose disease. Medical diagnosis is usually made by using some blood metrics and in addition, to be able to reach better results, one can benefit from different scientific methods. In this paper a Bayesian network method is proposed. This method is a hybrid that uses simple correlation and according to dependent variable type either simple linear regression or logistic regression for constructing a Bayesian topology. The Bayesian network is a method for representing probabilistic relationships between variables associated with an outcome of interest. To develop a Bayesian network, a structure must first be constructed. To build the topology of the Bayesian network, some alternative method can be used. One is using domain experts who usually have a good grasp of the conditional dependencies in the domain to develop the structure of the Bayesian network. Another is using structure learning algorithms, such as genetic algorithms, to construct the network topology from training data. In this paper a different construction method is proposed by using correlation analysis and one of the simple linear regression or logistic regression analyses. First, correlations of the examined variables are found. Then according to the significant correlation coefficients, the degree and direction of the interactions between these variables are established by using either simple linear regression or logistic regression. Finally the Bayesian network model is constructed by using this information. For evaluating our model, another model which does not have any relation between the input variables is also constructed. And these two models are compared by using an original thyroid data set. It is concluded that our proposed model provides a high degree of performance and good explanatory power and it may prove useful for clinicians in the medical field.