<![CDATA[Accurately capturing user motion intention is crucial for effective wrist control in myelectronic prosthetic hands. While various regression models have been explored to improve prediction performance, each presents specific limitations when used independently. This study proposes a novel ensemble learning approach that utilizes a Voting Regressor to combine the strengths of several regression models ANN, ANFIS, fuzzy logic, and their combinations (ANN-ANFIS, ANN-Fuzzy, ANFIS-Fuzzy, and ANN-ANFIS-Fuzzy) to improve predictive performance. Surface EMG signals were collected from the FCR and ECRL muscles at five contraction levels: 20%, 40%, 60%, 80%, and 100% MVC. These signals were used to predict wrist velocity, which was then validated using a SimMechanics based prosthetic hand model in MATLAB 2017a. The ensemble model outperformed all individual and combination models at four MVC levels; 20%, 40%, 60%, and 100%. However, at 80% MVC, a single model achieved superior performance. Based on the average performance gain at the four winning MVC levels, the ensemble method achieved an overall improvement of 11.38%. When applied to the prosthetic hand simulation, the ensemble model showed slight additional improvements in RMSE at each MVC level, highlighting the practical applicability of the approach. To assign optimal and objective weights to the contributing models, MCDM-WSM approach was applied. This method combined multiple evaluation metrics (RMSE, %NRMSE, MAE, R², and p-value) into a single composite score, leading to the final weighted regression equation: YVR-HG-wrist = (0.5163)YANN + (0.2367)YANFIS + (0.2470)YFuzzy. Furthermore, the ensemble model reduced reliance on additional control strategies such as PID tuning, as its improvements in RMSE were comparable to those typically achieved through PID-based compensation. These findings highlight the potential of a performance-weighted ensemble approach to provide more accurate, robust, and practical EMG-based prosthetic wrist control especially in real-time applications.]]>
