Heterogeneous Stacked Ensemble Framework for Surface Electromyography Signal Classification

Abstract

Surface electromyography (sEMG) signal is essential for accurately controlling prosthetic devices with numerous degrees of freedom in human-machine interfaces for robotics and assistive technologies. The controlling method of the upper-limb prosthesis device depends on electromyogram (EMG) pattern recognition, which requires the efficient blending of conventional signal processing and machine learning. This paper focuses on stacked ensemble models, one of the popular methods for reducing generalization error. The proposed work uses a dataset of sEMG signals from different upper-limb positions in subjects. The raw signals are transformed into correlated time-domain descriptors (cTDD) for feature extraction, which are then used to train the stacked ensemble framework. The framework includes four base classifiers (support vector machine (SVM), K-nearest neighbours (KNN), logistic regression (LR), and decision tree (DT)) and two meta-classifiers (random forest (RF) and multi-layer perceptrons (MLP). The performance of the meta-classifiers is evaluated on two test sets, showing superior classification accuracy compared to the basic classifiers. The proposed approach demonstrates the capability to accurately classify limb position invariant EMG signal classification for prosthetic device control.