Enhancing occupational safety in Construction: Predictive analytics using advanced ensemble machine learning algorithms

Abstract

In the construction industry, persistent occupational injury risks demand robust predictive methodologies to identify and mitigate hazards effectively. This study leverages advanced ensemble machine learning algorithms (Ml), Extreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), and a Modified Decision Tree (MDT) that evaluated against a comprehensive Occupational Safety and Health Administration (OSHA) dataset. Our findings reveal that the MDT, with tailored split criteria and advanced pruning techniques, achieved a prediction accuracy of 98.13 %, surpassing traditional models and other algorithms. XGBoost and LightGBM also showed robust performance with prediction accuracies of 92.9 % and 95.97 %, respectively. The efficacy of these models is demonstrated through k-fold cross-validation, with the MDT consistently outperforming others, recording a Mean Absolute Error (MAE) of 6.37, Mean Squared Error (MSE) of 59.57, and Mean Absolute Percentage Error (MAPE) of 7.97. These advancements underscore the potential of these models to enhance predictive accuracy in construction safety, enabling precise safety interventions to reduce risks and improve worker protection.