Cross Validation Configuration on k-NN for Finger Movements using EMG signals

It is already widely known that hand presence is very vital to perform daily activities successfully. However, it can be a different condition for people living with disabilities. They experience difficulty getting their activities done. Any tools that can replace hand movements are essential, such as hand prosthetics and hand exoskeleton. Nonetheless, the increasing accuracy score of hand motion classification using electromyography (EMG) signals must be further explored. This paper proposes a way to improve the accuracy of twelve finger movements classification task using k-Nearest Neighbor (k-NN) with two kinds of cross-validations, KFold and Stratified KFold evaluation. Before the EMG signal is fed to k-NN, this signal derived from four able-bodied subjects is extracted using Mean Absolute Value (MAV). Using k values of 1, 3, 5, and 7 on k-NN and shuffle and non-shuffle data on cross-validation, this study shows a comparative result of the above combination and directs future work. The shuffle data's accuracy rate outperforms the non-shuffle data with an accuracy of 89.97% compared to 65.39% on KFold and 90.24% compared to 82.08% on Stratified KFold. From this outcome, the shuffle data process significantly affects the accuracy level on hand movement classification.