How to find the casual relationship between variables using bayesian beleifnetwork in weka?

Question

How to find the causal relationship among variables from bayesian belief network?

All I understand is that, to find the causal relationship among variables, I need to construct Bayesian belief network.

I google a bit and all of the examples says that it requires expert knowledge and enough data to build a BBN.

I have seen there is an API in Weka for the bayesian network. But it is too complex for me. What would be step by step procedure in Weka so that I can build a Bayesian network from a dataset (dataset contains several variables and a target variable)

I have tried with the following code

    BayesNet bn= new BayesNet();
    //bn.m_Instances=data;
    LocalScoreSearchAlgorithm sa= new LocalScoreSearchAlgorithm();
    String[] options = new String[1];
    options[0] = "-D";                                    
    bn.setOptions(options);
    //bn.m_Instances=data;
    bn.initStructure();
    bn.initCPTs();
    bn.buildStructure();

but it did not work. I tried with several other combinations. But I am failed to build a network.

Once I build a network how to find the causal relationship among variables? How to interpret?

Answer 1

First of all, a Bayesian network learned from data only shows "correlation" among variables, not "causation". For instance if you have variables A, B, C, and D, and the network learned from data has A and B as parents of C, and C as the parent of D, it does not necessarily mean that A and B cause C, or that C causes D. It only means that they are correlated. Knowing causality among variables requires expert knowledge.

As of constructing a Bayesian network in Weka, you can simply use the GUI. When you open the GUI, click on "Explorer", then click on "Open file" and select your file (your file has to be in arff format). Then go on the "classify" tab. Click on "choose", and select "weka/classifiers/bayes/BayesNet". Then click on "start" and the network will be created.