Learning the Structure of Bayesian Networks: A Quantitative Assessment of the Effect of Different Algorithmic Schemes

Table 1

Summary of the major findings. The results of the Mann–Whitney U test [34] (one-tail test alternatives) in support of the results of our simulations are shown. The tests are performed on all the settings, with the comparisons as described in the first column of the table. We also show the considered metric, the adopted alternative of the one-tail test, the obtained value, and the mean of the two compared distributions of the results.

Comparison	Metric	Test	value	Mean 1	Mean 2
Density 0.4/0.8	Accuracy	Greater	<2.2e − 16	0.68	0.55
Nodes 10/15	Accuracy	Less	<2.2e − 16	0.61	0.62
Nodes 10/15	Hamming	Less	<2.2e − 16	17.40	39.54
Discr./contin.	Accuracy	Greater	<2.2e − 16	0.62	0.61
iamb/hc loglik	Sensitivity	Less	<2.2e − 16	0.02	0.59
iamb/hc loglik	Specificity	Greater	<2.2e − 16	0.98	0.38
ga/hc	Sensitivity	Less	<2.2e − 16	0.24	0.32
ga/hc	Specificity	Greater	<2.2e − 16	0.79	0.71