Finding Efficient Machine Learning Model for Hand Gesture Classification Using EMG Data
Adria Binte Habib, Faisal Bin Ashraf, Arif Shakil · 2021
Innovation in technology, increasing computing ability, and the invention of small devices have significantly influenced today’s life in different dimensions. Surface Electromyography (sEMG) records the electrical activity of skeleton muscles when it is placed on the hand. Hand gesture identification has become an important field of study in recent years especially controlling prostheses and other applications. So, the study of sEMG data has become pertinent as it can collect the electrical activity of muscles and help to find an efficient model that can identify different hand gestures efficiently. In this work, we have studied a number of traditional classification algorithms and deep learning classification techniques with a variety of parameters to find the most efficient model. Our findings disclose the fact that tree-based classification methods and LSTM work better for classifying EMG data. Considering all the factors, the Random Forest classifier is the optimal model and gives 99.43% accuracy with the lowest misclassification error. Moreover, LSTM gives 99.19% accuracy with a low misclassification error. Therefore, if we can train the model with a huge amount of data, Random forest can identify the pattern very well and outperform others. This model can be used in controlling devices like prostheses, digital wheelchairs, etc. with human activity.