Construction Hazard Prediction Based on Machine Learning Method

Construction accidents are a leading cause of worker fatalities worldwide. Identifying and addressing hazards is crucial to preventing these incidents, but effective prediction remains challenging due to limited data. This study proposes a method to create site safety inspection data for predicting construction hazards, employing machine learning for both single-hazard and compound-hazard prediction. Among 3,519 safety inspection cases, the Ensemble Bagged Trees model achieved 92.4% accuracy for single-hazard prediction, while Neural Networks reached 99% accuracy for predicting the top-5 most likely hazards. These models enable safety managers to proactively address hazards and reduce accident risks.