. 2021 Oct 19;26(20):6318. doi: 10.3390/molecules26206318

Table 4.

Model performance to compare classifiers (PLSDA against SVM) and different spectral regions for samples deposited on stainless steel.

	Spectral Region	LVs	Training Set (3218 Pixels)				Test Set (2864 Pixels)
	Spectral Region	LVs	OA	MCC	Sen	Spe	OA	MCC	Sen	Spe
PLSDA	4000–675 cm⁻¹	15	95%	0.90	0.99	0.91	90%	0.80	0.94	0.86
	1350–675 cm⁻¹	8	94%	0.89	0.99	0.90	88%	0.78	0.99	0.78
	1722–910 cm⁻¹	9	93%	0.87	0.95	0.92	89%	0.80	0.97	0.82
	3500–2600 cm⁻¹	10	99%	0.97	0.99	0.98	94%	0.89	0.94	0.95
SVM	4000–675 cm⁻¹	-	100%	0.99	0.99	1.00	94%	0.89	0.97	0.92
	1350–675 cm⁻¹	-	99%	0.98	0.99	0.98	94%	0.88	0.97	0.90
	1722–910 cm⁻¹	-	99%	0.97	0.99	0.98	94%	0.88	0.98	0.90
	3500–2600 cm⁻¹	-	99%	0.99	1.00	0.99	96%	0.93	0.96	0.97

OA: overall accuracy; MCC: Matthews correlation coefficient; Sen: sensitivity; Spe: specificity.