Feasibility of machine learning algorithms fed with frequency-domain features extracted from inertial signals to classify safe and unsafe postures during lifting tasks

The risk of developing work-related musculoskeletal disorders (WRMDs) is strongly correlated with the safe and unsafe postures during the weight lifting. Recently both the inertial signals - linear acceleration and angular velocity - acquired by wearable sensors and the Machine Learning (ML) techniques have shown to be able to successfully assess the biomechanical risk in the ergonomic filed [1]. The aim of this study was to assess the feasibility of four ML classifiers - fed with frequency-domain features extracted from inertial signals acquired by means of one inertial measurement unit (IMU) placed on the sternum - to classify safe and unsafe postures during weight lifting.