Channel and feature selection for a surface electromyographic pattern recognition task

• A novel variable selection method is proposed.
• 32 Electrode locations and 86 signal-features are evaluated for EMG control systems.
• We select 7 electrodes and 7 features that provide a high classification rate.
• Placing electrodes along the forearm is as important as around it.

The objective of this research is to select a reduced group of surface electromyographic (sEMG) channels and signal-features that is able to provide an accurate classification rate in a myoelectric control system for any user. To that end, the location of 32 sEMG electrodes placed around-along the forearm and 86 signal-features are evaluated simultaneously in a static-hand gesture classification task (14 different gestures). A novel multivariate variable selection filter method named mRMR-FCO is presented as part of the selection process. This process finds the most informative and least redundant combination of sEMG channels and signal-features among all the possible ones. The performance of the selected set of channels and signal-features is evaluated with a Support Vector Machine classifier.