Construction Safety Clash Detection: Identifying Safety Incompatibilities among Fundamental Attributes using Data Mining

• Published in: Automation in construction Vol. 74; pp. 39 - 54
• Main Authors: Tixier, Antoine J.-P., Hallowell, Matthew R., Rajagopalan, Balaji, Bowman, Dean
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.02.2017; Elsevier BV
• Subjects: Accidents; Advanced Work Packaging; AWP; BIM; Building Information Modeling; Construction accidents & safety; Construction methods; Data mining; Empirical analysis; Information Technology; Injuries; Machine learning; Natural language processing; Network analysis; Occupational safety; Personal communication; Prevention through design; Risk; Safety; Advanced Work Packaging; Network analysis; BIM; Machine learning; Prevention through design; Risk; AWP; Building Information Modeling; Safety; Information Technology
• ISSN: 0926-5805; 1872-7891
• DOI: 10.1016/j.autcon.2016.11.001

Construction still accounts for a disproportionate number of injuries, inducing consequent socioeconomic impacts. Despite recent attempts to improve construction safety by harnessing emerging technologies and intelligent systems, most frameworks still consider tasks and activities in isolation and use secondary, aggregated, or subjective data that prevent their widespread adoption. To address these limitations, we used a newly introduced conceptual framework and accompanying natural language processing system to extract standard information in the form of fundamental attributes from a set of 5298 raw accident reports. We then applied state-of-the-art data mining techniques to discover attribute combinations that contribute to injuries. We refer to these incompatibilities as “construction safety clashes”. The main contribution of our study lies in the methodological advancements that it brings to the construction safety domain. In light of the results obtained, our approach shows great promise to become a standard way of extracting valuable, actionable insights from injury reports in a fully unsupervised way. The use of our methodology could enable construction practitioners to ground their safety-related decisions on objective, empirical data, rather than on limited personal experience or expert opinion, which is the current industry standard. Finally, our methodology allows construction accidents to be viewed as perturbations in underlying networks of fundamental attributes. While the analysis of the current data set provides preliminary evidence for this theory, comparison to non-accident reports will be required for validation. •Concept of safety clashes is introduced•Safety risk attribute data sets are represented as weighted undirected graphs•Data mining is applied to graphs and attribute data to identify safety clashes•Qualitative analysis of the output shows promising potential of the methodology•The “accident as attribute network perturbations” theory is introduced