Machine Learning-Based Analysis of Occupational Cervical Pain in Bangladesh: Integrating Clinical and Non-clinical Determinants

Using machine learning methods, data obtained from 2815 Bangladeshi subjects indicated that approximately 42.5% of all participants experienced at least one type of musculoskeletal complaint. The rates of these complaints were significantly higher among female participants and those with a body mass index (BMI) classified as overweight and/or obese. Using principal component analysis (PCA) with a 3-dimensional biplot reduced the total number of variables and explained approximately 98% of the cumulative variance in the data. Among the different analytical prediction model types evaluated, the support vector machine (SVM) had the highest structural reliability (R2 = 0.82–0.90). As determined by SHAP interpretability analysis, BMI (20.3%) and age (18.1%) were significant clinical and non- clinical predictors, respectively. The use of random forest regression (RFR) demonstrated that individuals following either bioactive-rich or bioactive- poor standard diets, would experience significant reductions in cervical pain levels; however, individuals following a bioactive-rich diet were more likely to achieve these significant reductions than those following standard diets (p