Feature-channel subset selection for optimising myoelectric human-machine interface design

This paper proposes a feature-channel subset selection method for obtaining an optimal subset of features and channels of myoelectric human-machine interface applied to classify upper limb’s motions using multi-channel myoelectric signals. It employs a multi-objective genetic algorithm as a search strategy and either data separability index or classification rate as an objective function. A wide range of features in time, frequency, and time-scale domains are examined, and a modification that reduces the bias of cardinality in the separability index is evaluated. The proposed method produces a compact subset of features and channels, which results in high accuracy by linear classifiers without a need of preliminary tailor-made adjustments.