Information based feature selection for supervised motor unit action potential classification

The decomposition of a myoelectric (ME) signal into its constituent motor unit action potentials (MUAPs) can be considered as a classification problem. The choice of features used can affect the classifier performance. Using an information measure applied to clustering results the most discriminative features from a set of 32 time samples were selected. The full set of time samples, information-selected features, linear discriminant analysis and principle component analysis were used for the supervised classification of real MUAP data. Results suggest that the sets of information-selected features were an efficient representation of lower dimension which provided high accuracy classification with reduced computational requirements.