Abstract:

To explore the potential early warning value of construction safety hazard data and improve the efficiency of hazard ide.pngication and control, this study investigated a data-driven approach for the association analysis and early warning strategy of construction safety hazards by integrating text mining, association rule mining, and complex network theory. First, 1 405 construction safety inspection records in 2023 were standardized and dimensionally reduced using text mining techniques, resulting in the extraction of 67 safety hazard features. Then, 70 frequent itemsets and 125 strong association rules were obtained using the Apriori algorithm, and the types of hazard associations were ide.pngied. Subsequently, a hazard feature network was constructed based on complex network theory. Key hazard features were ide.pngied through structural and node-level indicators, combined with feature modularity analysis. Finally, a feature-driven early warning strategy was proposed. The results show that text mining effectively reduces the dimensionality of unstructured hazard data. The hazard feature network based on association rules successfully reveals hidden associations within the data and enhances the reliability of early warning information, providing clear direction and reference for on-site hazard detection. The early warning strategy helps address the issues of disorder and inefficiency in traditional hazard inspections.

Key words: data-driven, construction, safety hazards, association rules, warning strategy, complex networks