Table 5.

Comparison of the classifier level fusion methods based on SVM (C = 10${}^{-1}$, C = 10${}^{-3}$) and k-NN (k = 1, k = 3) for mental workload estimation. 6 pairs of mental workload conditions (1-back VS 2-back, 1-back VS 3-back, 2-back VS 3-back, 1 and 2-back VS 3-back, 1-back VS 2 and 3-back, 1-back VS 2-back VS 3-back) are used to compare the ability of the methods to classify the mental workload levels in 1-, 2-, 3-back tasks. VGG-MKL is a multi-kernel learning algorithm based on SVM.

Conditions	Methods	k-NN (%)		SVM (%)
		k = 1	k = 3	C = 10${}^{-1}$	C = 10${}^{-3}$
1-back VS. 2-back	Average	53.0 ± 13.2	54.0 ± 12.6	54.0 ± 25.7	56.1 ± 29.1
	VGG-MKL	−	−	56.0 ± 17.4	55.3 ± 20.1
	LCDM	75.5 ± 19.0	75.4 ± 19.5	79.2 ± 17.0	81.1 ± 18.4
	IMIM-C	85.0 ± 7.7	87.7 ± 7.9	91.9 ± 11.6	97.1 ± 3.0
1-back VS. 3-back	Average	62.1 ± 14.7	61.9 ± 15.7	47.8 ± 26.0	49.1 ± 23.6
	VGG-MKL	−	−	45.8 ± 22.5	43.8 ± 21.4
	LCDM	66.1 ± 18.1	66.3 ± 14.1	78.1 ± 18.5	78.5 ± 18.1
	IMIM-C	89.8 ± 10.6	90.4 ± 9.8	96.4 ± 7.0	99.6 ± 0.9
2-back VS. 3-back	Average	57.5 ± 12.2	59.1 ± 13.9	48.6 ± 14.3	48.3 ± 12.5
	VGG-MKL	−	−	45.7 ± 28.3	44.2 ± 26.0
	LCDM	64.1 ± 18.5	70.4 ± 18.9	67.4 ± 12.3	66.9 ± 12.2
	IMIM-C	77.6 ± 9.8	79.2 ± 8.9	72.6 ± 39.1	86.6 ± 13.2
1-back VS. 2-back, 3-back	Average	60.8 ± 5.0	60.6 ± 4.3	64.1 ± 19.0	65.6 ± 18.4
	VGG-MKL	−	−	55.6 ± 21.1	53.9 ± 21.1
	LCDM	71.2 ± 14.5	70.3 ± 13.5	78.6 ± 16.1	79.9 ± 18.7
	IMIM-C	90.1 ± 5.4	90.9 ± 4.9	98.0 ± 3.8	98.8 ± 1.9
1-back, 2-back VS. 3-back	Average	65.0 ± 10.9	65.2 ± 11.6	47.8 ± 24.0	49.2 ± 21.9
	VGG-MKL	−	−	48.7 ± 14.5	48.1 ± 12.9
	LCDM	64.6 ± 10.9	70.9 ± 18.0	70.6 ± 11.8	71.8 ± 9.8
	IMIM-C	81.8 ± 9.6	83.1 ± 8.8	82.0 ± 36.3	91.1 ± 8.8
1-back VS. 2-back VS. 3-back	Average	43.4 ± 3.1	43.6 ± 3.9	34.3 ± 23.8	35.4 ± 22.5
	VGG-MKL	−	−	28.3 ± 18.6	26.3 ± 19.2
	LCDM	53.0 ± 2.5	56.2 ± 2.7	61.0 ± 3.3	63.0 ± 3.5
	IMIM-C	76.6 ± 10.8	78.5 ± 9.6	80.2 ± 25.4	89.9 ± 8.6