Feasibility of Tree-based Machine Learning algorithms fed with surface electromyographic features to discriminate risk classes according to NIOSH