Assessing Biomechanical Risk using Tree-Based Machine Learning Algorithms fed with Postural Sway Parameters derived from Angular Velocity Data

Work-related musculoskeletal disorders affect soft tissues, including muscles, tendons, and ligaments, and often arise from workplace physical demands such as poor posture, repetitive tasks, and mechanical strain. These conditions are major contributors to absenteeism, disability, and reduced productivity. Traditional ergonomic methods, like the Revised NIOSH Lifting Equation (RNLE), provide a framework for evaluating biomechanical risks and preventing potential musculoskeletal injuries. However, these methods can be complex and time-intensive to implement. Advances in wearable sensors and artificial intelligence offer promising alternatives for more efficient biomechanical risk assessment. This study explored the performances of machine learning models in classifying biomechanical risk levels, as defined by the RNLE. The models were trained using postural sway parameters derived from angular velocity data, captured using a single inertial measurement unit placed on the lumbar region, corresponding to the body’s center of mass. Eight volunteers participated in the study protocol involving two lifting sessions, each assigned a predefined risk level calculated using the RNLE. Then, four tree-based machine learning models fed with sway parameters derived from angular velocity data were implemented. The results demonstrated excellent performances in classifying biomechanical risk classes, with Gradient Boosted Tree achieving the highest accuracy of 95.4% and an impressive area under the receiver operating characteristic curve of 99.0%. Although the limited sample size and demographic imbalances necessitate further investigation, the findings indicate that the proposed methodology holds strong potential in occupational ergonomics for biomechanical risk assessment associated with weight lifting.