This study assesses artificial intelligence-based control algorithms for a transradial myoelectric prosthesis. The analysis is supported by a dataset collected from 20 Peruvian participants with transradial amputation, including congenital and traumatic cases. Each participant performed 240 gesture repetitions under varying postures with surface electromyographic (sEMG) signals recorded on the user’s forearm. The dataset was processed to extract time and frequency domain features, enabling the implementation of classifiers such as Neural Networks (NN), Random Forest (RF), Extreme Gradient Boosting (XGB), and Decision Trees (DT). The results demonstrate that RF and XGB outperformed other classifiers when employed in a stack model architecture (97.4% accuracy). Distal amputations exhibited superior outcomes, as results from users of congenital amputations were also superior. Favorable results were also observed among individuals with a medium time since limb loss (26–51 years). Initial tests on a Raspberry Pi Zero 2 W system validated the feasibility of real-time implementation with reduced sliding window sizes. These findings highlight the potential of bespoke machine learning approaches to enhance gesture recognition accuracy, contributing to the development of affordable, personalized prosthetic solutions for individuals with transradial amputations.
García et al. (2026) studied this question.