Table 2.

Best classification rates over all manifold learning methods and classifiers with the use of the Euclidean distance $L_2$ (see “Construction of FCN based on the Euclidean distance” section); parameters, PT, classifier, accuracy (Acc), sensitivity (Sens) and specificity (Spec) rates. Classifiers are noted as RSVM (Radial SVM), LSVM (Linear SVM), k-NN (k-NN Classifier) and ANN (Artificial Neural Networks)

Method	Parameters	PT	Classifier	Acc ± SD (%)	Sens (%)	Spec (%)
MDS	p = 3	0.54	RSVM	72.3 ± 1.7	52.8	83.4
		0.36	LSVM	57.9 ± 2.5	26.6	79.6
		0.26	k-NN	65.7 ± 2.2	59.6	66.1
		0.44	ANN	64.9 ± 2.6	53.7	69.5
ISOMAP	p = 2, k = 5	0.68	RSVM	72.9 ± 2	55.8	81.9
		0.42	LSVM	54.6 ± 0.1	0	95.9
		0.68	k-NN	65 ± 2.3	47.7	74.7
		0.66	ANN	68.2 ± 1.8	47.4	80.4
Diffusion maps	p = 5, $\sigma$ = 110	0.66	RSVM	68.8 ± 2.2	57.7	73.1
		0.52	LSVM	62.9 ± 1.9	56.7	63.5
		0.26	k-NN	65.1 ± 2.5	63.3	61.7
		0.5	ANN	63.9 ± 2.6	71.4	53.2
kPCA	p = 5, $\gamma$ = 11.5	0.62	RSVM	67.1 ± 1.3	37.1	87.1
		0.28	LSVM	62.1 ± 1.6	56.8	62.9
		0.64	k-NN	62.7 ± 2.8	56.7	63.3
		0.62	ANN	67.2 ± 1.6	62.1	66.8
LLE	p = 3, k = 3	0.48	RSVM	70.3 ± 2.6	67.9	67.3
		0.26	LSVM	70.2 ± 1.4	68.9	66.4
		0.28	k-NN	65.3 ± 2.7	59.4	65.4
		0.26	ANN	70 ± 2	64.3	70
Euclidean matrix	–	0.64	RSVM	71.6 ± 1.7	60.1	76.1
		0.58	LSVM	59.2 ± 2.5	62.5	52.3
		0.64	k-NN	72 ± 2.2	67.5	70.6
		0.70	ANN	62.1 ± 2.6	60.4	59.1