Skip to main content

Table 4 Correlations between classification accuracy and nearest neighbor separability

From: Classification complexity in myoelectric pattern recognition

 

Average result

Individual results

 
 

LDA (AMI)

Single/OVO

MLP AMI/MIX

SVM AMI/MIX

LDA (AMI)

Single/OVO

MLP AMI/MIX

SVM AMI/MIX

Data set

K = 20

0.86/0.98

0.96/0.97

0.90/0.90

0.83/0.93

0.92/0.92

0.72/0.72

SM

0.86/0.97

0.98/NA

0.74/NA

0.84/0.92

0.97/NA

0.70/NA

IM

K = 120

0.90/0.97

0.92/0.95

0.92/0.92

0.87/0.90

0.87/0.89

0.73/0.73

SM

0.90/0.98

0.97/NA

0.78/NA

0.89/0.93

0.94/NA

0.73/NA

IM

  1. The correlation between classification accuracy and NNS with different values of the parameter k. Correlations under “individual results” were calculated using classification accuracies and NNS from every individual movement, subject and feature, while those under “average result” were derived using one average NNS and classification accuracy for every subject and feature. Both methods provide one correlation, although “individual results” use more data. Classifiers were configured using AMI or MIX. Classifiers were used in the conventional “single” topology, apart from LDA, which was used in “single” and OVO. The highest correlation values per column are highlighted in bold. All correlations were found statistically significant at p < 0.01. The MIX configuration is not applicable (NA) for individual movements since there are no mixed outputs