Logo image
Back
Explainable Ensemble Machine Learning for Predicting Injury Severity in Agricultural Accidents
Journal article   Peer reviewed

Explainable Ensemble Machine Learning for Predicting Injury Severity in Agricultural Accidents

Omer Mermer, Eddie Zhang and Ibrahim Demir
Journal of agromedicine
04/23/2026
DOI: 10.1080/1059924X.2026.2658048
PMID: 42026846

View Online

Abstract

Objectives Agricultural injuries remain a leading occupational hazard, causing substantial human and economic losses worldwide. This study aimed to predict agricultural injury severity using linear and ensemble machine learning (ML) models, while ensuring interpretability through explainable artificial intelligence (XAI). Methods We analyzed 2,421 agricultural incidents (2015–2024) from AgInjuryNews, categorized as fatal or non-fatal. Data were pre-processed to remove duplicates, outliers, and incomplete records, then split into training and testing sets. Multiple ML models – including Naïve Bayes, Decision Tree, Support Vector Machine, Random Forest, and Gradient Boosting – were trained and optimized using cross-validation. Ensemble approaches (bagging, boosting, stacking, and voting) were also implemented. Model performance was evaluated using accuracy, precision, recall, and F1-score. Shapley Additive Explanations (SHAP) were applied to identify key predictors of injury severity. Results Ensemble models achieved the best overall performance, with Random Forest, XGBoost, and LightGBM outperforming linear classifiers. XGBoost achieved near-perfect recall for fatal injuries, though classification of non-fatal cases remained challenging due to class imbalance. SHAP analysis consistently identified age, gender, location, and time of incident as the most influential predictors across models. These findings highlight both the predictive power of ensemble methods and the value of XAI in understanding underlying risk factors. Conclusion Ensemble ML approaches, supported by explainable AI techniques, offer effective tools for predicting injury severity in agriculture and uncovering critical contributing factors. The results underscore the importance of addressing data imbalance for improved classification of non-fatal injuries. Insights into key demographic and environmental predictors can inform targeted safety interventions and policy development, contributing to reduced injury rates and enhanced protection for agricultural workers.
 Machine Learning Agricultural injury ensemble models explainable AI (XAI) injury severity prediction SHAP analysis

Details

Metrics

1 Record Views
Logo image