Related Experiment Videos
Explainable Ensemble Machine Learning for Predicting Injury Severity in Agricultural Accidents
Omer Mermer1, Eddie Zhang2, Ibrahim Demir1,3
1By Water Institute, Tulane University, New Orleans, LA, USA.
Journal of Agromedicine
|April 24, 2026
Summary
Machine learning models accurately predict agricultural injury severity. Explainable AI identified age, gender, location, and time as key risk factors for farm injuries.
Area of Science:
- Agricultural Safety
- Occupational Health
- Machine Learning Applications
Background:
- Agricultural injuries pose significant global risks, impacting human well-being and economic stability.
- Predictive modeling can enhance understanding and mitigation of these hazards.
Purpose of the Study:
- To predict agricultural injury severity using machine learning (ML) models.
- To ensure model interpretability via explainable artificial intelligence (XAI).
Main Methods:
- Analysis of 2,421 agricultural incidents (2015-2024) from AgInjuryNews.
- Implementation and evaluation of various ML and ensemble models (e.g., Random Forest, XGBoost).
- Application of Shapley Additive Explanations (SHAP) for predictor identification.
Main Results:
- Ensemble models, particularly XGBoost, demonstrated superior performance in predicting injury severity.
- XGBoost achieved high recall for fatal injuries, though non-fatal injury classification faced challenges due to data imbalance.
- SHAP analysis identified age, gender, location, and time as critical predictors.
Conclusions:
- Ensemble ML and XAI provide effective tools for predicting agricultural injury severity and identifying risk factors.
- Addressing data imbalance is crucial for improving non-fatal injury prediction.
- Findings can inform targeted safety interventions and policy development to reduce agricultural injuries.