Construction accident data mining: A retrospective study using structural equation modeling based on 10-year data

Abstract

BACKGROUND: The data mining of construction accidents based on a robust modeling process can be used as a practical technique for reducing the frequency of construction accidents. OBJECTIVE: This study was designed to data-mine construction accidents. METHODS: This study was conducted in 2020 on construction accidents in Iran for ten years (2009–2018). The instruments to collect the required data were the checklists and descriptive reports of the accidents. The dependent variables of the study included reactive safety indicators related to construction accidents (lost working days (LWD) and total accident costs (TAC)). The independent variables consisted of four latent factors: personal variables, organizational variables, unsafe working conditions, and unsafe acts. The data were collected based on the conceptual model designed for data mining. The data mining process was carried out based on the structural equation modeling by IBM AMOS V. 23.0. RESULTS: A total of 5742 construction accidents occurring in 10 years were analyzed. The means of TAC and LWD indicators were estimated to be 248.20±52.60 days and 1893.10±152.22 $. These two indicators directly correlated with the two latent factors of unsafe conditions and unsafe acts and their related variables and were indirectly influenced by latent personal and organizational factors. The relationship between unsafe conditions and unsafe acts was significantly positive. The relationship between latent personal and organizational factors and the two construction accident indicators was significantly negative (p <0.05). CONCLUSION: The model results showed that personal and organizational variables could, directly and indirectly, affect reactive safety indicators in construction projects. Thus, these findings can be used to design and improve safety strategies to prevent and decrease construction accidents and incidents.