Bagged regression trees for simultaneous myoelectric force estimation

A novel application of bootstrap aggregating (bagged) regression trees is proposed for simultaneous force estimation of multiple degrees of freedom (DOFs). Ten able-bodied subjects participated and wrist flexion-extension, abduction-adduction, and pronation-supination were investigated (data from the work of Ameri et al., 2013). The estimation accuracies were compared to those of the widely used multilayer perceptron artificial neural networks (ANNs). The bagged trees outperformed the baseline ANNs, slightly but significantly, in abduction-adduction (p<;0.05), while for flexion-extension and pronation-supination DOFs, no significant difference was found (p>0.1) between the bagged tress and ANNs. The results suggest that bagged regression trees can be an alternative approach for potential use in simultaneous myoelectric control.