Examining the Impact of Different K Values on the Performance of Multiple Algorithms in K-Fold Cross-Validation

Myoelectric interface advancements have the potential to modify the use of wearable prosthetics as limb replacements by using EMG signals to control active hand and arm prostheses. Accurately identifying separate motions is crucial for a myoelectric control system to function effectively. Cross-validation is an efficient strategy that can aid in myoelectric control by making sure a machine learning model generalizes to different users and different kinds of movements effectively. The k-fold cross-validation’s k value is an important factor that can significantly affect the efficiency of various classification models. Therefore, the goal of this study was to determine the k-fold cross validation's most precise number of folds that can increase the sensitivity and effectiveness of motion recognition using EMG data. Thus, in this study, the performance of four machine and deep learning models as Support vector machine (SVM), Artificial neural networks (ANN), Decision tree (DT) and K-nearest neighbors (KNN) was examined in relation to various k values (2–20) during cross-validation to predict different wrist movements. The training performance of the classification models was evaluated using two popular evaluation metrics i.e. mean classification accuracy (MCA) and area under the curve (AUC). Out of all four classifiers, KNN consistently exhibited the highest accuracy across all k values. The inconsistent performance results demonstrated that for the same classification problem, different classification algorithms perform differently depending on the value of k. In some cases, increasing the value of k will result in better accuracy, while other times it will merely result in more computational complexity. Therefore, it is recommended to carry out trials with multiple values of k with a specific dataset and model to find the best value of k for improving the accuracy.