An Artificial Neural Network Classifier for palm Motion categorization based on EMG signal

Electromyogram (EMG) signals have become more prevalent in recent years for the purpose of hand and finger motion identification. On the other hand, the majority of research have concentrated their attention on the arm and the entire hand, rather than on individual finger (IF) motions, which were thought to be more challenging. EMG-based classifications for hand and finger gestures are being developed using data mining algorithms in this study. Constant circuit arrangement is the basis for these algorithms. Ten individuals in good health were asked to make ten different hand/finger gestures, seven of which were IF movements. Three channels' worth of Electromyogram (EMG) signals was measured, and then each channel's worth of data was broken down into six time-domain (TD) characteristics. Artificial neural networks (ANN), a support vector machine, a random forest (RF) as well as a logistic regression were all used to create a set of 10 distinct gesture-specific classifications. There were a total of 18 qualities to choose from. For example, SVMs achieved mean precision of 0.840; ANNs were 0.840; SVMs were 0.866; SVMs were 0.877; and SVMs were 0.831.Each analysis of deviation and F-tests showed that now the artificial neural network (ANN) had the highest mean accuracy and the least inter-subject variation in efficiency, suggesting that individual variability in EMG data had the least influence on the ANN. We obtained a larger ratio of motions to channels than previous research that were comparable to ours using solely TD characteristics, which suggests that the suggested technique has the potential to increase the system's usability and minimize the amount of computing work required.