Application of Ensemble Machine Learning for Construction Safety Risk Assessment

The rising prevalence of fatalities in the construction industry has prompted management to shift from traditional approaches to more advanced methods for analysis like machine learning (ML). Each construction project must undergo a risk assessment to understand the safety status of their construction sites and adopt preventative measures to avoid catastrophic incidents. The purpose of this study is to develop a prediction model for risk assessment of construction sites using ensemble machine learning techniques. A dataset from the Occupational Safety and Health Administration database of 4847 event reports from 2015 to 2017 is used for the analysis. Firstly, the primary risk factors causing accidents are identified and were divided into four: Before Accident, After Accident, Critical Factors-1 (CR-1), and Critical Factors-2 (CR-2). Using these attribute sets, predictive models were generated with five different classifiers with the help of different methods of resampling. The analysis was executed using both simple and ensemble modelling and the latter showed better results. The best performing classifiers under each attribute set were identified. Among the different models, the Gradient Boosting model trained with a CR-2 set of attributes, exhibited the best prediction results. Throughout the study, the application of ML in safety management has proven to be effective. The predictive model developed assists the safety management team comprehend the safety status of their specific construction projects and as a result, adopt appropriate preventive measures.