Skip to content

Classification complexity in myoelectric pattern recognition

By Kirstin Ahmed

Depending on a pre-processing step known as feature extraction, an EMG classifier can have better or worse discrimination capabilities with respect to its classes. We explore what metrics can be used to indicate the effectiveness of different feature extraction strategies. We find that nearest neighbour separability (NNS) and separability index (SI) computed on the extracted features correlate strongly to the performance of final classification.