TABLE 1.

Disorder–order classification using three wavelengths. ^a

	Wavelength (nm)				Error (%)
Cutoff (nm)	Algorithm	WL1	WL2	WL3	Ordered	Disordered	Global
175	SVM–RBF	182	194	209	0	0	0
	Discr-quadratic	179	214	225	0.8	0	0.7
	Tree-medium	192	220	228	0.8	0	0.7
180	KNN-fine	184	197	208	0.7	0	0.6
	Discr-quadratic	197	216	221	1.3	0	1.1
	SVM–RBF	195	217	227	2	0	1.7
	Tree-simple	185	192	211	2	0	1.7
185	Tree-medium	191	201	250	1.3	2.7	1.6
	SVM–RBF	195	217	227	2	2.4	2.1
	Discr-quadratic	199	213	234	2	2.4	2.1
190	Tree-medium	191	201	250	1.9	5.6	2.8
	SVM–RBF	196	216	229	2.4	5.1	3.1
	Discr-quadratic	199	213	234	3.5	1.7	3.1
195	Discr-quadratic	199	213	234	3.5	2.9	3.3
	SVM-linear	196	212	235	4.1	1.5	3.4
	KNN-cosine	197	206	233	4.7	1.5	3.8
	SVM–RBF	196	216	223	3.5	4.4	3.8
	Discr-linear	195	219	237	3.5	4.5	3.8
200	KNN-cosine	212	217	225	4.7	1.5	3.8
	SVM-linear	202	205	231	7.2	2.5	5.7
	SVM–RBF	206	212	229	5	7.5	5.7
	Discr-quadratic	201	211	215	6.6	3.8	5.7
	KNN-fine	212	215	227	3.9	10	5.7
205	KNN-cosine	212	217	225	3.3	7.4	4.6
	SVM–RBF	206	212	229	5	7.4	5.7
	KNN-fine	212	215	227	3.9	9.9	5.7

^aAlgorithms showing the least errors using three wavelengths (WL1, WL2, WL3) for classification as a function of the cutoff wavelength are presented. For training dataset, for a given wavelength triplet, all proteins’ spectra that covered those wavelengths were used.