Download PDFOpen PDF in browserClassification of Fatal and Non-fatal Construction Incidents in the Southeastern U.S. using Machine Learning10 pages•Published: July 23, 2025AbstractConstruction sites face persistent safety challenges, with incidents often resulting in severe injuries or fatalities. In the U.S., these safety concerns are heightened due to the high volume of construction projects and complex working conditions. This study conducts a data-driven analysis of construction safety incidents in the Southeastern U.S., utilizing five Machine Learning (ML) techniques to classify fatal or non-fatal incidents. A dataset of 1,963 incidents obtained from the OSHA was analyzed with the ML techniques for their prediction accuracy of classifications. Key findings reveal that random forest and decision trees achieved the highest accuracy and reliability in classifying fatal or non-fatal incidents, with random forest outperforming all models in the classifications. Feature importance analysis highlighted factors such as age, height, occupation, and event type as significant predictors of injury severity. The study’s implications are substantial for construction safety management; ML models can provide predictive insights that support proactive safety measures on construction sites. By identifying high-risk factors associated with severe injuries, this research contributes to the development of data-driven safety interventions and policy improvements aimed at reducing incident rates. The findings underscore the potential of ML in advancing construction safety through targeted risk assessment and preventive strategies.Keyphrases: construction safety, data driven analysis, machine learning, risk prediction In: Wesley Collins, Anthony J. Perrenoud and John Posillico (editors). Proceedings of Associated Schools of Construction 61st Annual International Conference, vol 6, pages 856-865.
|
