I have fitted several Bayesian network models to complete training data with different local probabilistic distribution but with a fixed network structure. I want to choose only one of them. How can I go about this? I've read that I can choose the model which gives the lowest prediction error or other loss metric, as well as some goodness of fit criteria like some information criterion. But how can I tell which of these approaches are better? Thanks!
Check out these papers:
"Scoring functions for learning Bayesian networks" by Alexandra M. Carvalho, http://www.lx.it.pt/~asmc/pub/talks/09-TA/ta_pres.pdf
"Comparison of Score Metrics for Bayesian Network learning" by Shulin Yang and Kuo-Chu Chang, http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.93.511&rep=rep1&type=pdf
"Metrics for evaluating performance and uncertainty of Bayesian network models" by Bruce G. Marcot, http://www.fs.fed.us/pnw/pubs/journals/pnw_2012_marcot002.pdf?