Learning mixed graphical models with separate sparsity parameters and stability-based model selection

BMC Bioinformatics

Table 1 Comparison of model selection methods

Methods	Precision	Recall	F1-score	Matthews CC	Accuracy
AIC	0.1104 (0.002)	0.9698 (0.004)	0.1982 (0.003)	0.2882 (0.003)	0.7952 (0.003)
BIC	0.4588 (0.028)	0.8633 (0.007)	0.5890 (0.025)	0.6098 (0.022)	0.9652 (0.004)
CV	0.1530 (0.003)	0.9694 (0.004)	0.2640 (0.005)	0.3539 (0.004)	0.8587 (0.003)
Oracle	0.9149 (0.015)	0.7868 (0.021)	0.8397 (0.009)	0.8416 (0.008)	0.9923 (0.000)
StARS – 1 λ	0.8988 (0.018)	0.4993 (0.010)	0.6408 (0.011)	0.6632 (0.011)	0.9854 (0.001)
StEPS – 3 λ	0.9159 (0.014)	0.6720 (0.009)	0.7731 (0.007)	0.7787 (0.007)	0.9897 (0.000)

AIC akaike information criterion, BIC Bayesian information criterion, CV cross-validation, Oracle: best possible prediction performance (maximize accuracy using true graph)
Mean (and standard error) of classification performance over 20 datasets simulated from scale-free networks

ISSN: 1471-2105