. 2020 Oct 6;10:16581. doi: 10.1038/s41598-020-73644-6

Table 2.

Three top performing binary classifiers for HemoPI-1, HemoPI-2, HemoPI-3 datasets after optimizing classifier hyperparameters and number N of descriptors.

Classifiers		Hyperparameters	Feature reduction	N	Acc. (%)	Prec. (%)	MCC statistic	CK statistic	AUC ROC
HemoPI-1 model and validation datasets
1.1	LDA	'solver': 'svd', 'tol': 0.0001	RFECV	18	95.1	92.6	0.903	0.903	0.951
1.1	LDA	'solver': 'svd', 'tol': 0.0001	RFECV	18	94.6	92.5	0.891	0.891	0.946
1.2	GBC	'max_depth': 4, 'max_features': 'sqrt', 'min_samples_leaf': 10, 'n_estimators': 240	MC (0.75)	26	96.5	95.0	0.930	0.930	0.965
1.2	GBC		MC (0.75)	26	92.7	89.6	0.855	0.855	0.927
1.3	GBC	'max_depth': 4, 'max_features': 'sqrt', 'min_samples_leaf': 10, 'n_estimators': 208	None	56	96.0	94.6	0.921	0.921	0.960
1.3	GBC		None	56	92.3	89.2	0.846	0.846	0.923
HemoPI-2 model and validation datasets
2.1	GBC	'max_depth': 4, 'max_features': 'sqrt', 'min_samples_leaf': 2, 'n_estimators': 112	None	56	77.7	74.0	0.549	0.549	0.774
2.1	GBC		None	56	74.3	70.4	0.479	0.476	0.736
2.2	GBC	Default	RFECV	15	77.8	74.2	0.552	0.552	0.775
2.2	GBC	Default	RFECV	15	73.2	69.8	0.459	0.482	0.728
2.3	GBC	Default	None	56	76.7	72.9	0.529	0.528	0.763
2.3	GBC	Default	None	56	72.3	68.9	0.439	0.437	0.717
HemoPI-3 model and validation datasets
3.1	GBC	'max_depth': 18, 'max_features': 'log2', 'min_samples_leaf': 10, 'n_estimators':192	None	56	80.0	76.4	0.597	0.597	0.796
3.1	GBC		None	56	71.7	68.3	0.427	0.425	0.711
3.2	GBC	'max_depth': 12, 'max_features': 'sqrt', 'min_samples_leaf': 8, 'n_estimators': 160	RFECV	40	78.2	74.4	0.559	0.558	0.777
3.2	GBC		RFECV	40	74.5	70.8	0.483	0.482	0.740
3.3	GBC	'max_depth': 20, 'max_features': 'log2', 'min_samples_leaf': 8, 'n_estimators': 96	MC (0.75)	28	78.0	74.4	0.556	0.556	0.777
3.3	GBC		MC (0.75)	28	72.6	69.0	0.445	0.443	0.719

Optimal number N of descriptors were determined using multicollinearity, RFECV: tenfold cross-validated recursive feature extraction or BE: backward elimination.