Discovering workers' actions leading to severe construction accidents using accident report data and sequence mining techniques

Historical construction accident reports have been widely used to gain insights into the primary causes of past incidents in construction. Previous studies have successfully identified accident causes and affected body parts. However, there remains a gap in understanding the high-risk actions of workers. This study aims to fill this gap by conducting a novel investigation into the most prevalent sequential patterns between workers' actions prior to severe accidents. The study extracted sequential accident patterns by applying the PrefixSpan sequential pattern mining algorithm on a large database of action-accident-consequences manually built from the Occupational Safety and Health Administration's construction accident reports. Social Network Analysis was then performed to determine high-risk workers' actions leading to severe accidents. Additionally, statistical tests were employed to explore the sectoral differences in the rank of high-risk actions. The study revealed the priority for 24 high-risk actions leading to severe accidents in construction. The ranking of these actions was found statistically different between construction sectors. Organizations can utilize the findings to develop targeted safety programs and interventions to mitigate future incidents in the construction industry. This study, however, was limited by the size of the sequential database, resulting from the manual data annotation process. This issue could be mitigated in future research by exploring semiautomated annotation approaches.