Thus multi-position classifiers were trained only using
EMG data from multiple arm positions (P1-P5) and tested
using EMG data from all five arm positions. The average
classification errors of 31 movement classifiers (26 multiple
positions plus 5 single position) from the amputated arms
of all five subjects were calculated and are presented in
Figure 6(a). For the sake of comparison, the average error
bars from subjects’ intact arms were also calculated and
are plotted in Figure 6(a). Generally speaking, the average
classification error gradually decreased along with more
arm positions including in classifier training. When the
EMG data from all five arm positions were involved in the
training set, the average classification error reached a minimum
value of around 10.8% for the amputated arm and
10.3% for the intact arms, as shown in Figure 6(b). It is
noteworthy that in most of the 31 motion classifiers, the