Predictive Risk Modeling via Natural Language Processing of Industrial Safety Reports

Rapid management of industrial safety is producing long incident investigation reports, which are not fully reflected in the use of text. The study examines Natural Language Processing (NLP) and Machine Learning (ML) in transforming qualitative messages on safety into quantitative predictive messages. A total of 16,878 records of construction accidents were used to test different ML algorithms in terms of usefulness and Elasticity. The model of the Fields of Application was the Random Forest with an accuracy of 79.3%, 77.1% precision, 78.0% recall, and an Area Under the Receiver Operating Curve (AUROC) of 0.98. The analysis of the importance of the features revealed that accident mechanism and nature were mainly important predictors, whereas the temporal and economic factors were the least affected. These results support the effectiveness of NLP when using unstructured safety data, which provides practitioners with a resource-based and proactive risk intervention tool as opposed to a reactive mechanism