Comparing the evaluation measures for the optimal networks learned from the Austra datasets with different sizes. In this figure, we compare the performance of the four evaluation metrics (SHD, AHD, accuracy, and sensitivity) for the Australian Credit Approval dataset. The y-axis label indicates which evaluation metric that graph displays. We display the results for α = 1 for BDeu for all measures because it had the best convergence behavior for this dataset. We used the behavior of each of the curves to evaluate the convergence of the corresponding scoring function. We consider a scoring function to have converged for an evaluation metric when increasing the dataset size does not change the value for that scoring function and evaluation metric. Thus, we look for "flat lines" in the graphs.
Liu et al. BMC Bioinformatics 2012 13(Suppl 15):S14 doi:10.1186/1471-2105-13-S15-S14