Feasibility and challenges in AI-driven risk assessment of musculoskeletal disorders (MSDs) in office workers

Although musculoskeletal disorders are a major concern for office workers, traditional risk assessment tools do not sufficiently capture the complex, non-linear interactions among physical, psychosocial, and individual factors. To address the above, this study aimed to provide a more nuanced, data-driven assessment of MSD risks. A dataset of 810 office workers, integrating four publicly available sources, was used to train and optimise six machine learning models [Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), eXtreme Gradient Boosting (XGB), and Multilayer Perceptron (MLP)] for predicting MSDs across nine body regions. External validation was conducted using an independent dataset of 52 office workers. Tree-based ensemble models (GB and XGB) achieved the strongest predictive performance, particularly for neck, shoulders, upper back, and lower back disorders, confirming the non-linear nature of MSD risk. However, model performance declined during external validation, underscoring challenges in generalisability and the need for context-specific tuning. Feature importance analyses revealed body-region-specific risk profiles shaped by different weights of contributions of individual, physical, and psychosocial factors. Interestingly, laptop/phone use showed no predictive importance, reflecting its ubiquity rather than risk absence. This study demonstrates the feasibility and value of AI-driven ergonomic risk assessment. By generating body-part-specific risk profiles, the approach enables targeted, holistic interventions that integrate biomechanical, psychosocial, and individual health factors. Despite challenges in generalisability, these findings establish a foundation for transforming office ergonomics into a proactive, data-driven practice.