Table 5.

Coffee bean classification models for species identification (Arabica-Robusta) based on Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) for green and roasted whole coffee beans. Sample size = 510 for green; 340 for roasted coffee.

Classification method	Function	Pre-treatment	Green		Roasted
			Correct (%)	Incorrect (%)	Correct (%)	Incorrect (%)
LDA	Linear	Log (1/R)	98.39	1.61	98.53	1.47
		1st derivative	99.29	0.71	100.00	0.00
		SNV	98.57	1.43	98.53	1.47
		MSC	98.75	1.25	98.53	1.47
		2nd derivative	99.61	0.39	100.00	0.00
	Quadratic	Log (1/R)	100.00	0.00	100.00	0.00
		1st derivative	99.80	0.20	99.12	0.88
		SNV	99.80	0.20	99.71	0.29
		MSC	100.00	0.00	99.41	0.59
		2nd derivative	99.80	0.20	98.82	1.18
Classification method	Function	Pre-treatment	Training accuracy (%)	Validation accuracy (%)	Training accuracy (%)	Validation accuracy (%)
SVM	Polynomial	Log (1/R)	74.90	74.12	87.06	84.42
		1st derivative	72.55	72.55	64.71	64.71
		SNV	88.43	87.06	94.12	91.76
		MSC	82.75	82.16	95.00	88.24
		2nd derivative	72.55	72.55	64.71	64.71
	Radial basis function	Log (1/R)	84.90	83.92	90.29	88.24
		1st derivative	91.18	91.18	88.24	88.24
		SNV	97.45	97.25	98.23	97.06
		MSC	94.51	94.12	95.00	94.12
		2nd derivative	89.02	88.82	84.71	84.42