EMG-based pattern recognition approach in post stroke robot-aided rehabilitation: a feasibility study

Table 7 GROUP II: classification results and confusion matrices for TEST3

Patient id and classification rate	Actual classes	Predicted classes
		North	East	South	West
P1	North	66.7	20	25	14.3
53.3%	East	16.7	40	41.7	0
	South	0	20	33.3	0
	West	16.7	20	0	85.7
P2	North	100	0	0	0
70%	East	0	87.5	0	0
	South	0	0	27.3	0
	West	0	12.5	72.7	100
P3	North	66.7	25	0	0
36.7%	East	16.7	75	75	50
	South	0	0	0	0
	West	16.7	0	25	50
P4	North	28.6	12.5	0	12.5
36.7%	East	57.1	37.5	14.3	12.5
	South	14.3	50	57.1	50
	West	0	0	28.6	25
P5	North	25	14.3	0	33.3
43.3%	East	37.5	57.1	16.7	0
	South	12.5	28.6	83.3	44.4
	West	25	0	0	22.2
P6	North	87.5	0	0	0
83.3%	East	12.5	71.4	12.5	0
	South	0	28.6	75	0
	West	0	0	12.5	100
P7	North	100	0	0	14.3
70%	East	0	50	55.6	14.3
70%	South	0	50	44	0
	West	0	0	0	71.4

Elements on the left-right diagonal indicate the percentage of correct classification, while elements on the off-diagonals denote percentage of misclassification. SVM was trained (70% of trials) and tested (30% of trials) individually on each patient with the muscle dataset recorded during the experimental session (see Methods); training and testing data were randomly selected and mutually exclusive; test was repeated 5 times for each patient and accuracy rates were averaged across iterations.

ISSN: 1743-0003