Exploiting Experts' Knowledge for Structure Learning of Bayesian Networks.

Area of Science:

• Computer Science
• Statistics
• Machine Learning

Background:

• Learning Bayesian network structures from observational data is computationally challenging due to the vast search space.
• Observational data alone cannot uniquely identify the true causal graph, as multiple graphs may explain the same independencies.

Purpose of the Study:

• To investigate if incorporating domain expert opinions can improve Bayesian network structure learning.
• To develop new scoring functions that leverage expert knowledge on cause-effect relationships, considering varying expert accuracies.

Main Methods:

• Developed two novel scoring functions integrating expert opinions and observational data.
• Modeled expert accuracy using three parameters.
• Employed an expectation-maximization algorithm to estimate expert accuracies for the first scoring function.
• Marginalized out accuracy parameters in the second scoring function for robustness.

Main Results:

• Experimental results on simulated and real-world datasets demonstrate improved structure learning.
• Exploiting expert knowledge, particularly when considering their accuracy, leads to better causal discovery.

Conclusions:

• Incorporating domain expertise, with careful consideration of individual expert reliability, enhances Bayesian network structure learning.
• The proposed scoring functions offer effective methods for integrating heterogeneous expert knowledge into causal discovery.