Proactive construction hazard prevention model using machine learning

Construction accidents are a major cause of occupational fatalities worldwide. While effective hazard prediction is crucial for preventing accidents, it remains challenging due to the lack of available data. This study introduces the proactive construction hazard prevention (PCHP) model, combining a construction hazard causation hierarchy (CHCH), a machine learning-based hazard prediction model (HPMB), a lessons learned file (LLF), and a proactive hazard prevention (PHP) module. Unlike traditional accident prediction models, the proposed PCHP model connects worksite attributes with lessons learned from previous hazard prevention cases, forming a proactive approach to hazard prevention that is unprecedented in the literature. The proposed model was tested on 5,071 hazard records from 114 construction sites. It demonstrated 92.3% accuracy for single-hazard predictions using the ensemble bagged trees mode, and 96.8% accuracy for the top five compound-hazard predictions using neural networks. This research contributes to advance construction safety management by: (1) validating machine learning's effectiveness in hazard prediction with unprecedented accuracy; (2) developing a novel method for predicting complex compound hazards previously unexplored in the literature; and (3) introducing a proactive model for implementing preventive measures before construction begins.