Analysis of construction site accidents for preventive measures using natural language processing techniques

In the construction industry, workplace safety is a major concern. Across all sectors, construction workers are more likely injured compared to other industry due to lack of safety and awareness. Accidents at construction sites cause enormous financial loss in addition to human misery. Analysis of accidents is crucial for developing scientific risk control strategies and preventing the recurrence of similar mishaps in the future. Accident reports give summaries and causes of the related events. The processing and analysis of these reports in order to take preventive action has made extensive use of Machine Learning and Natural Language Processing (NLP). In the proposed work, these approaches are applied to the analysis of construction accident records. Support Vector Machines (SVM), Linear Regression (LR), K-Nearest Neighbour (KNN), Decision Trees (DT), Naive Bayes (NB), and Ensemble models have been proposed for categorising the causes of the accidents. In this study, the optimized ensemble model performs more effectively than the other models in terms of average weighted F1 score.