Machine Learning-Driven EMG-Based 3D Prosthetic Arm with 5 Degrees of Freedom

the study includes the designing, testing and controlling of prosthetic arm, which is flexible, wearable, non-invasive, efficient and cost-effective. It has 5 degrees of freedom and can perform six gestures. It was developed under 500 USD which is a low cost, has a mass of approximately 1 kg and was tested on ten subjects. The surface electromyogram (sEMG) signals were obtained from four MyoWare EMG sensors attached to amputee's forearm detecting electrical activity in four muscles: Brachialis, Palmaris longus, Flexor digitorum, and Extensor digitorum. Using raw sEMG signals, six time-domain features namely integrated signal, zero crossing, variance, root mean square, mean absolute value, and short time Fourier transform were calculated and considered for classification. The processed features were then fed into different three classifiers namely decision tree (DT), support vector machine (SVM), and artificial neural networks (ANN) for the purpose of training the classifiers. It was also deduced that root mean square and short time fourier transform as features and SVM as classifier performed relatively better with (p= 0.04) and accuracy 92.35 percent as compared to all other features and classifiers. In the offline mode the classification accuracy is 85.18,81.16 and 92.35 percent for ANN, DT and SVM classifiers respectively. Similarly, in the real mode, the least individual accuracy is 85 percent. To ensure that the results were accurate, all the classifiers and the vibrotactile feedback mechanism were put to the test. In the case of the vibrotactile force sensing feedback mechanism, statistical testing revealed a p-value of 0.04 and a correlation coefficient of 0.9985 between matching and target forces exerted by the fingers, disproving the null hypothesis and establishing an authentic connection.