Determining critical combinations of safety fatality causes using spectral clustering and computational data mining algorithms

Although the Occupational Safety and Health Administration (OSHA) has played a key role in decreasing safety accidents on construction sites, such reduction is not well perceived among the different types of incidents. For instance, while the injuries resulting in days away from work were substantially decreased, construction fatalities rose in the previous years. While previous safety fatality research work concentrated on studying how different variables could influence the number of fatalities in the construction industry, there is a lack of research efforts that studied the associations, interdependencies, and critical combinations between the different fatality causes. As such, to solve the critical fatality problem that the construction industry is facing, more advanced management methods should be applied. That said, this paper aims to address this important knowledge gap through a methodology based on data-driven computational algorithms. The data analyzed in this paper is comprised of 100 OSHA national fatal accident case files that are prepared by OSHA compliance officers. First, a reference matrix of 100 case files was constructed based on 60 causes that have shown to be the most pertinent to construction fatalities based on a previous study. Second, the spectral clustering computational algorithm was used to cluster the fatality causes based on the strength of their interconnectivities. Third, the frequent pattern data mining and Apriori algorithms were used to determine the critical combinations and associations of causes that mostly contribute to fatal accidents on construction sites. The findings indicate that the 60 fatality causes could be categorized into five clusters. Thus, the most critical combination of fatality causes was determined within each one of the identified clusters. Also, the findings reflected that while safety accidents could happen due to factors or causes that are individually critical, fatalities on construction sites could also result due to a combination of factors that might not be perceived to be critical on the individual level but rather become critical when combined with other factors. This study adds to the body of knowledge by proposing a data-driven accident causation approach to analyze construction fatality accidents, which adds a different dimension to the traditional analysis of safety accidents by factoring possible associations and combinations between different causes. That said, this paper equips safety professionals with a proactive approach that allows them to take the needed preventive actions to avoid fatalities on construction sites by identifying, in hindsight, the critical combinations and associations of fatality causes. Ultimately, the outcomes of this paper would enhance the safety performance in the construction industry and prevent construction fatalities.