Biomechanical risk evaluation through machine learning algorithms fed with features extracted from sEMG of neck extensors

Work-related musculoskeletal disorders (WRMDs) affect millions of workers worldwide, posing substantial economic burdens on industries and healthcare systems. Prolonged exposure, repetitive tasks, awkward postures and intensive efforts are keys factors contributing to the development of WRMDs. Several quantitative or semiquantitative methodologies are employed to evaluate the biomechanical risk and to prevent WRMDs in the occupational ergonomics field. However, these methods are still timeconsuming and operator-dependent. Recently, the application of wearable sensors combined with artificial intelligence is providing remarkable results in terms of biomechanical risk assessment in the occupational ergonomics field. Therefore, in the present work, we examined the potential of Machine Learning (ML) models to differentiate between biomechanical risk categories as defined by the Revised NIOSH Lifting Equation (RNLE). The ML models were trained using timedomain and frequency-domain features extracted from surface electromyographic (sEMG) signals obtained from the neck extensor muscles of four healthy subjects during weight-lifting tasks. The study findings indicated that the Support Vector Machine algorithm performed the best, achieving an accuracy of 83.6% and an area under the receiver operating characteristic curve of 89.9%. However, the study was limited by its small sample size and the restricted age range of the volunteers. Future research involving a larger and more diverse population in terms of age and number of subjects could further validate the effectiveness of the proposed methodology.