Integrated Machine Learning Analysis for Investigating Injury Factors Among Korean Farmers

This study aimed to identify key risk and protective factors for agricultural injuries by integrating machine learning–based feature-importance analyses with traditional statistical tests. We analyzed nationally representative survey data from 17,770 Korean farmers using an integrated analytical pipeline combining ensemble machine learning, interpretability techniques (feature importance, permutation importance, and SHAP), and statistical testing. Recurrent injuries were strongly linked to high-risk tasks such as ladder use, heavy material handling, pesticide spraying, and the operation of passenger machinery. Higher injury rates among farmers perceiving agriculture as “very dangerous” highlighted a strong association between risk awareness and actual injury experience. Demographic and socioeconomic factors, male, farm ownership, and lower income, were likewise confirmed as significant predictors of injury risk. These findings demonstrate the value of data-driven approaches and emphasize the need for targeted safety training, provision of protective equipment and machinery support, as well as continued monitoring of vulnerable farming households to guide policy interventions.