Improving the Robustness of Myoelectric Control System Using Linear Regression Classifier

In the application of electromyography-based pattern recognition (EMG-PR) approach, the stability and robustness of EMG-PR control system against EMG description methods like feature extraction and electrode configuration was scarcely investigated. Aiming at developing a robust, stable, and accurate EMG-PR system, a new pattern recognition method, Linear Regression Classifier (LRC) is proposed in this study. The results among 12 TBI patients show that the proposed LRC scheme achieved significantly higher average classification accuracy (achieved above 99% when using 56 monopolar electrodes) in comparison to the commonly used LDA and KNN. Moreover, the LRC scheme was robust to the selection of feature sets, and to the electrode configurations especially when TD and TDAR feature sets were used. These outcomes suggest the comparative advantage of the LRC.