Abstract

Upper limb amputees are individuals who lost their hands due to trauma and injury. Controlled prosthesis based on surface Electromyography (sEMG) signals recovers the lost functionality for upper limb amputees. Several pattern recognition techniques help amputees in controlling prosthesis by classifying different upper limb movements intuitively. The proposed framework performs an analysis of classification of upper limb movements on real time and retrieved surface Electromyography (sEMG) signal data. Band pass filter is used in pre-processing stage and Time Domain features are extracted. The Features selection analysis is also performed wherein Extra Tree classifier and histogrambased features is used for retrieved and real time data respectively. The pre-processed real time and retrieved data with features and classes are fed to the classification stage. The hyperparameters of the classifiers are tuned using Grid Search Method. The classifiers to be stacked are Adaptive Boosting, Gradient Boosting Machine, Quadratic Discriminant Analysis, Linear Discriminant Analysis, K Nearest Neighbor and Random Forest. The properties of the proposed stacking classifier are diverse and same error rate classifiers procured using McNemar's hypothesis testing. The evaluation metrics considered are Accuracy, Precision, Recall and F1 Score. The evaluation results signify that stacking classifier provides a highest accuracy in all experiments.