Modeling Finger Movements of a Prosthetic Hand from EEG Signals Using ARX SIMO Models

This study presents a comparative modeling analysis of a prosthetic hand controlled by electroencephalographic (EEG) signals. A Single Input Multiple Output (SIMO) framework is adopted, where a single EEG signal is used to predict the movements of five fingers (thumb, index, middle, ring, and little finger). Four dynamic models are evaluated: Auto-Regressive with eXogenous input (ARX), Auto-Regressive Moving Average with eXogenous input (ARMAX), Box–Jenkins (BJ), and Transfer Function (TF) models. Performance is assessed using Root Mean Square Error (RMSE), coefficient of determination (R²), robustness index (1/RMSE), and ANOVA statistical analysis. Results demonstrate that the ARX model achieves the best overall performance, with a low global RMSE of 28.6782 and a high global R² of 0.7784 . Finger-wise analysis confirms the superiority of the ARX model, particularly for the thumb and index fingers. ANOVA results further validate the statistical significance of EEG influence on most finger movements, confirming the feasibility of real-time EEG-based prosthetic hand control.