Exploring associations between accident types and activities in construction using natural language processing

Abstract

Existing accident analysis methods often fail to integrate the relationship of construction accidents with construction activities, restrict the potential for visualizing the accidents frequency in Building Information Modelling (BIM). This study proposes a decision support system that provides visual insights, and the frequency of the core construction accidents in relation to typical construction activities. K-mean clustering and keyword-matching techniques are used to group the leading causes of construction accidents and then compare them with the construction activities. The average hit rate computed for construction accidents was 91%, whereas the hit rate calculated for construction activities connected to these frequent accidents was 75%. The classification models are integrated into the Power BI platform to offer decision-makers deeper insights regarding the relationship of prevalent construction accident types and associated activities. The practical application of the system is demonstrated with a case study, exemplifying the fall accident data integrated in a BIM environment.