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

Many work activities can imply a biomechanical overload. Among these activities, lifting loads may determine work-related musculoskeletal disorders. In order to limit injuries, the National Institute for Occupational Safety and Health (NIOSH) proposed a methodology for assessing biomechanical risk in lifting tasks by means of a math formula based on intensity, duration, frequency and other geometrical characteristic of the lifting. In this work, we explored the feasibility of tree-based machine learning algorithms to classify biomechanical risk according to the Revised NIOSH lifting equation. Electromyography signals acquired from the biceps during lifting loads were collected using a wearable sensors for surface electromyography on a study population composed of 5 healthy young subjects. The EMG signals were segmented in order to extract the region of interest related to the lifting actions and for each region of interest several features in time and frequency domains were extracted. High results - greater than 95% - were obtained in terms of evaluation metrics for a binary risk/no-risk classification. In conclusion, this work indicates the proposed combination of features and machine learning algorithms represents a valid approach to automatically classify risk activities according to the Revised NISOH lifting equation. Future investigation on enriched study population could confirm the potentiality of this methodology to automatically classify potential risky activities.