Table 10.

Results obtained in simulated data set TIED. “MB” stands for “Markov boundary”, and “VS” stands for “variable set”. The 95% interval for weighted accuracy denotes the range in which weighted accuracies of 95% of the extracted Markov boundaries/variable sets fell. Classification performance of the MAP-BN classifier in the same data sample was 0.966 weighted accuracy. Highlighted in bold are results that are statistically comparable to the MAP-BN classification performance.

Method		I. Number of distinct MBs or VSs	II. Average size of extracted distinct MBs or VSs	III. Number of true MBs identified exactly	IV. Average proportion of false positives	V. Average false negative rate	VI. Weighted accuracy over all extracted MBs or VSs
							Average	95% Interval
TIE*	max-k = 3, α = 0.05	72	5.0	72	0.000	0.000	0.951	0.938	0.965
iTIE*	max-k = 3, α = 0.05	72	5.0	72	0.000	0.000	0.951	0.938	0.965
KIAMB	Number of runs = 5000, α = 0.05, K = 0.7	377	2.8	0	0.000	0.400	0.727	0.479	0.946
	Number of runs = 5000, α = 0.05, K = 0.8	377	2.8	0	0.000	0.400	0.727	0.479	0.946
	Number of runs = 5000, α = 0.05, K = 0.9	377	2.8	0	0.000	0.400	0.727	0.479	0.946
EGS-NCMIGS	l = 7, δ = 0.015	6	7.0	0	0.286	0.000	0.964	0.963	0.965
	l = 7, K = 10	6	10.0	0	0.500	0.000	0.964	0.963	0.965
	l = 7, K = 50	6	21.0	0	0.762	0.000	0.941	0.937	0.943
	l = 5000, δ = 0.015	24	7.3	0	0.469	0.267	0.954	0.843	0.967
	l = 5000, K = 10	20	10.0	0	0.610	0.220	0.964	0.954	0.970
	l = 5000, K = 50	9	21.0	0	0.762	0.000	0.944	0.937	0.954
EGS-CMIM	l = 7, K = 10	6	10.0	0	0.500	0.000	0.963	0.963	0.965
	l = 7, K = 50	6	21.0	0	0.762	0.000	0.939	0.937	0.942
	l = 5000, K = 10	20	10.0	0	0.595	0.190	0.963	0.951	0.969
	l = 5000, K = 50	9	21.0	0	0.762	0.000	0.943	0.937	0.954
EGSG	Number of Markov boundaries = 30, t = 5	30	7.0	0	0.476	0.267	0.840	0.605	0.968
	Number of Markov boundaries = 30, t = 10	30	7.0	0	0.548	0.367	0.722	0.379	0.962
	Number of Markov boundaries = 30, 1 = 15	30	7.0	0	0.548	0.367	0.722	0.379	0.962
	Number of Markov boundaries = 5,000, t = 5	1,997	7.0	0	0.286	0.000	0.863	0.620	0.965
	Number of Markov boundaries = 5,000, t = 10	3,027	7.0	0	0.286	0.000	0.774	0.500	0.965
	Number of Markov boundaries = 5,000, t = 15	3,027	7.0	0	0.286	0.000	0.774	0.500	0.965
Resampling+RFE	without statistical comparison	1,374	14.9	1	0.397	0.058	0.955	0.932	0.979
	with statistical comparison (α = 0.05)	188	4.9	0	0.171	0.378	0.930	0.917	0.967
Resampling+UAF	without statistical comparison	184	20.8	0	0.752	0.000	0.953	0.934	0.966
	with statistical comparison (α = 0.05)	19	8.4	0	0.592	0.347	0.930	0.917	0.938
IR-HITON-PC	max-k = 3, α = 0.05	3	4.3	1	0.083	0.200	0.946	0.936	0.965
IR-SPLR	without statistical comparison	1	26.0	0	0.808	0.000	0.958	0.958	0.958
	with statistical comparison (α = 0.05)	1	17.0	0	0.706	0.000	0.959	0.959	0.959