Optimization of a support vector regression model predicts maximal weight lifted during a manual material handling assessment within a military setting

Manual materiel handling (MMH) is an important task within the military and routine physical assessments ensure personnel are proficient in MMH tasks. Research demonstrates that predictive models can accurately estimate MMH performance. Optimization of these models may enhance military workflows by providing insights into maximal military performance of MMH tasks. This study recruited a mixed population of 51 general and military participants to assess physical characteristics and MMH lift-to-place performance. Previously developed models (support vector regressions (SVR) and ridge) were retrained and optimized through hyperparameter and feature selection tuning based on physical characteristic metrics. Root mean square errors (RMSE) evaluated the model's prediction performance. The best retrained SVR models had lower prediction error than the best retrained ridge models (RMSEs of 3.04 and 3.62 kgs vs 3.32 and 4.24 kgs, respectively). SVR models require minimal implementation effort and provide robust predictive capabilities, making them a valuable tool for military organizations. • Developed models can predict the maximum weight lifted by military personnel. • Completely retraining models optimizes generalizability in military populations. • Support vector regressions outperform ridge models for lift-to-place prediction. • Optimized models may reduce the burden of personnel task selection.