A Two-Layer Weighted Ensemble Voting Approach for EMG-Based Prosthetic Wrist Control for Finger Pinch

Human motion intention is a critical factor that significantly influences a prosthetic hand control design. Motivated by this limitation, this study proposes an improved approach for prosthetic wrist movement control during finger pinch using EMG signals. This work highlights a limitation using a single model regression technique by integrating these based models (ANN, ANFIS, Fuzzy, ANN-ANFIS, ANN-Fuzzy, ANFIS-Fuzzy, ANN-ANFIS-Fuzzy) into two layer ensemble framework. Two muscles FCR and ECRL were analysed under different wrist position (neutral, flexion and extension). This proposed framework utilised performance-based weighting approach using multiple evaluation metrics (RMSE, %NRMSE, MAE, R2 and p-value) followed by MCDM-WSM to determine the optimal weightage stacking configuration. Experimental results show that ensemble combination outperformed individual models with ANN-ANFIS-Fuzzy achieve the best performance for FCR (RMSE=45.28°/s) and ANFIS-Fuzzy for ECRL (RMSE=35.61°/s). Additional trajectory stabilization and fluctuation reduction were observed following PID implementation. The final stacked model shows an improvement in robustness, reliability and predictive performance velocity results. These preliminary findings demonstrate the possibility of ensemble learning in facing the nonlinearity issues in EMG signals while contributing more accurate and prediction performance under controlled experimental conditions.