Work-related musculoskeletal disorders (WMSDs) of the upper limbs remain a significant concern in occupational health, and their assessment is generally based on subjective reports, limiting the consistency, scalability, and accuracy of ergonomic screening. This study proposes a methodological approach integrating Item Response Theory (IRT) and Machine Learning (ML) to enable automated classification of upper-limb musculoskeletal discomfort levels among workers. The sample consisted of 300 individuals assessed using a reliable and validated psychometric instrument (McDonald’s omega = 0.94; all items showed factor loadings and commonalities greater than 0.50). The IRT parameters enabled the construction of a hierarchical musculoskeletal discomfort scale comprising six levels, anchored in cumulative response probabilities and ranging from minimal to maximal discomfort. Six ML algorithms were trained to estimate individual discomfort levels based on IRT scale levels and WMSD symptoms. The Support Vector Machine (SVM) achieved the best performance, with an F1-score above 96% and low variability confirmed through 5,000 bootstrap resamples (standard deviation < 4%). Shapley Additive Explanations (SHAP) analysis identified symptoms in the forearm and hand regions as key predictors of the outcome. The results demonstrate the feasibility of IRT–ML integration for automated ergonomic risk screening, reducing subjectivity and enabling robust, scalable, individualized analysis through combined psychometric and predictive power. This psychometric-computational approach functions as a data-driven surrogate expert system, embedding expert knowledge within a validated measurement model and operationalizing it through machine learning for scalable, individualized risk screening in Industry 4.0 environments.
Silva et al. (Sun,) studied this question.