Comparative analysis of classical and ensemble models for predicting whole body vibration induced lumbar spine stress. A case study of agricultural tractor operators

Accurate prediction of lumbar health is necessary for developing effective ergonomic strategies for tractor operators exposed to whole-body vibration. This study aims to predict static compression dose (S ed ), a key measure of lumbar spine stress as per ISO 2631-5, by comparing classical regression and ensemble models. Three tractor operation parameters (average speed, average depth, and pulling force) are considered to assess S ed during rotary tillage operation. The performance of two classical models (Linear and Huber regression) is compared with five ensemble models (Random Forest, Gradient Boosting, XGBoost, AdaBoost, and Bagging regressors) in predicting S ed . The comparison identifies the best models in each category, with linear regression achieving a mean bootstrap R 2 of 0.91 (95 % CI: 0.87 to 0.94) and Random Forest achieving 0.93 (95 % CI: 0.90 to 0.95). To further enhance performance, meta-models are developed using two meta-learners (Random Forest and Gradient Boosting) to integrate classical and ensemble models. These models are optimized using different ensemble strategies: simple averaging, weighted averaging, stacking, and voting regressors. Among these, the stacking method proves most effective, achieving a mean bootstrap R 2 of 0.94 (95 % CI: 0.93 to 0.96). Feature importance analysis reveals that the multi-model combination of ensemble models achieves the highest predictive score (0.99) for S ed . These findings demonstrate that ensemble models outperform classical models in predicting Sed, particularly when combined through stacking methods. This advancement has significant implications for improving occupational health and safety among tractor operators, potentially leading to better ergonomic tractor designs aimed at reducing lumbar spine stress. • Comparing classical models (Linear and Huber Regression) with ensemble models (Random Forest, Gradient Boosting, XGBoost, AdaBoost, and Bagging Regressors) to predict S ed . • Applying advanced meta-modeling strategies to improve prediction performance. • Using Explainable AI techniques, such as feature importance analysis, to better understand operational parameters affecting S ed and to ensure model interpretation.