Table 5.

Twofold cross‐validation results for the two pixel‐level bleeding detection datasets. The performance measures are presented in the original/ flipped order regarding the selected training and test sets.

Color features	Texture features	Classifier	PREC	REC	SPEC	ACC	F1	MCC	ROC	PRC
None	All	RT	0.882/0.851	0.875/0.854	0.762/0.745	0.875/0.854	0.878/0.852	0.607/0.620	0.818/0.800	0.850/0.800
All	None	RT	0.936/0.921	0.934/0.922	0.873/0.864	0.934/0.922	0.935/0.922	0.787/0.800	0.903/0.893	0.914/0.887
All	All	RT	0.958/0.943	0.957/0.944	0.911/0.898	0.957/0.944	0.957/0.943	0.859/0.856	0.934/0.921	0.941/0.916
All	Five top	RT	0.959/0.944	0.958/0.944	0.918/0.896	0.958/0.944	0.958/0.944	0.863/0.857	0.938/0.920	0.943/0.916
All	Five different	RT	0.958/0.946	0.958/0.946	0.912/0.903	0.958/0.946	0.958/0.946	0.861/0.862	0.935/0.925	0.941/0.920
All	All	RF	0.971/0.957	0.970/0.957	0.946/0.924	0.970/0.957	0.971/0.957	0.903/0.890	0.993/0.985	0.992/0.985
All	Five top	RF	0.970/0.955	0.969/0.955	0.945/0.919	0.969/0.955	0.969/0.955	0.899/0.886	0.992/0.984	0.990/0.984
All	Five different	RF	0.971/0.956	0.970/0.956	0.945/0.922	0.970/0.956	0.970/0.956	0.902/0.888	0.993/0.984	0.991/0.984
All	All	LMT	0.968/0.955	0.968/0.956	0.940/0.923	0.968/0.956	0.968/0.955	0.894/0.887	0.989/0.982	0.984/0.983
All	Five top	LMT	0.969/0.954	0.968/0.955	0.944/0.918	0.968/0.955	0.968/0.954	0.896/0.884	0.988/0.982	0.979/0.981
All	Five different	LMT	0.968/0.954	0.967/0.954	0.939/0.920	0.967/0.954	0.967/0.954	0.893/0.883	0.988/0.983	0.977/0.983
‐	‐	ZeroR	0.665/0.533	0.816/0.730	0.184/0.270	0.816/0.730	0.733/0.616	0.000/0.000	0.500/0.500	0.699/0.605

PREC, precision; REC, sensitivity or recall; SPEC, specificity; F1, F‐Measure; ACC, accuracy; MCC, Matthews correlation coefficient; ROC, receiver operator characteristic curve; PRC, precision‐recall curve; RT, Random tree; LMT, logistic model tree.