Integrated Explainable Ensemble Machine Learning Prediction of Injury Severity in Agricultural Accidents

Agricultural injuries remain a significant occupational hazard, causing substantial human and economic losses worldwide. This study investigates the prediction of agricultural injury severity using both linear and ensemble machine learning (ML) models and applies explainable AI (XAI) techniques to understand the contribution of input features. Data from AgInjuryNews (2015–2024) was preprocessed to extract relevant attributes such as location, time, age, and safety measures. The dataset comprised 2,421 incidents categorized as fatal or non-fatal. Various ML models, including Naïve Bayes (NB), Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (GB), were trained and evaluated using standard performance metrics. Ensemble models demonstrated superior accuracy and recall compared to linear models, with XGBoost achieving a recall of 100% for fatal injuries. However, all models faced challenges in predicting non-fatal injuries due to class imbalance. SHAP analysis provided insights into feature importance, with age, gender, location, and time emerging as the most influential predictors across models. This research highlights the effectiveness of ensemble ML models in injury prediction while emphasizing the need for balanced datasets and XAI techniques for actionable insights. The findings have practical implications for enhancing agricultural safety and guiding policy interventions.