Predicting construction workers’ safety participation behaviour using explainable machine learning

Abstract

Purpose The performance of the construction industry is hampered by safety issues arising from hazardous working conditions, including accidents and injuries, often linked to unsafe worker behaviours. While safety behaviour is based on safety compliance and safety participation. This study employs machine learning to develop a model to predict safety participation behaviour. Design/methodology/approach A comparative machine learning framework using eight classification algorithms was employed to identify key behavioural, cognitive and organisational determinants influencing safety participation behaviour. Findings Random Forest model achieved superior performance with 87.72% accuracy and 84.21% after tuning, significantly outperforming other methods. Subsequent model interpretability analyses using SHAP values and partial dependence plots identified safety motivation, safety attitudes and the application of safety knowledge as forming a tripartite foundation for safety participation behaviour. These findings demonstrate how predictive analytics can serve as a diagnostic and decision-support tool within safety management. Originality/value Using a comparative machine learning approach, this study evaluated the determinants of safety participation behaviour among construction workers. It provides predictive analytics to support safety management by enabling the timely identification of workers exhibiting high levels of unsafe behaviour, enabling proactive interventions.