Table 2.

Comparison of different hybrid approaches using cross-validation for the benchmark data sets.

	PS1302
Localization	PSL101		PSLseq+PSL101		PSLsse+PSL101
	Acc (%)	MCC	Acc (%)	MCC	Acc (%)	MCC
CP	97.2 (94.8)	0.91 (0.89)	96.4 (94.4)	0.90 (0.89)	95.6 (94.4)	0.90 (0.90)
IM	94.4 (92.9)	0.95 (0.94)	93.3 (91.8)	0.95 (0.93)	93.3 (91.8)	0.94 (0.93)
PP	87.7 (88.1)	0.86 (0.84)	88.9 (88.9)	0.86 (0.85)	91.4 (91.0)	0.88 (0.88)
OM	94.3 (93.8)	0.94 (0.91)	95.5 (95.7)	0.96 (0.93)	96.3 (96.9)	0.96 (0.95)
EC	87.9 (83.2)	0.87 (0.84)	89.5 (85.8)	0.89 (0.87)	90.0 (87.9)	0.89 (0.89)
Overall	92.7 (91.2)	-	93.1 (91.9)	-	93.7 (92.9)	-
	PS1444
Localization	PSL101		PSLseq+PSL101		PSLsse+PSL101
	Acc (%)	MCC	Acc (%)	MCC	Acc (%)	MCC
CP	96.0 (94.2)	0.91 (0.90)	94.6 (92.8)	0.89 (0.88)	95.0 (93.5)	0.91 (0.90)
IM	94.5 (92.6)	0.95 (0.94)	93.5 (91.6)	0.94 (0.93)	93.5 (91.6)	0.94 (0.93)
PP	85.1 (88.0)	0.82 (0.83)	87.0 (88.4)	0.84 (0.83)	90.2 (91.7)	0.86 (0.87)
OM	94.9 (93.9)	0.93 (0.91)	95.9 (95.7)	0.95 (0.93)	96.7 (96.4)	0.96 (0.95)
EC	82.6 (83.2)	0.83 (0.85)	87.9 (86.3)	0.87 (0.88)	87.4 (87.9)	0.87 (0.89)
Overall	91.6 (91.1)	-	92.4 (91.6)	-	93.2 (92.8)	-

§ The performance of incorporating a three-way data split procedure is indicated in the parentheses.