Title Automated Extraction of Construction Safety Accident Patterns via Multi-Rule Combination Based on Association Rule Mining
Authors 박영준(Park, Young-Jun)
DOI https://doi.org/10.5659/JAIK.2026.42.3.387
Page pp.387-395
ISSN 2733-6247
Keywords Construction Accidents; Association Rule Mining; Multi-Rule Combination; Accident Pattern Analysis; Safety Management
Abstract Construction accidents are characterized by the complex interaction of multiple factors, including work processes, work locations, construction trades, and safety measures. Due to this complexity, conventional accident analysis approaches that focus on single factors have inherent limitations. To address these limitations, this study applies Association Rule Mining (ARM) to construction accident data and proposes an automated multi-rule?based accident pattern derivation process that systematically combines and summarizes multiple association rules by accident type. First, accident data items not aligned with the analysis objectives were identified and removed, and similar items were integrated into higher-level categories to construct a categorical data-set. Association rules were then extracted, and meaningful rule sets were identified through the stepwise application of support, confidence, and lift thresholds reflecting the sparse and heterogeneous nature of construction accident data. The filtered rules were automatically grouped according to common accident outcomes, and composite accident patterns were quantitatively derived using the number of rules, average support, average confidence, and average and maximum lift values. The analysis results showed that cut and laceration accidents exhibited a single-concentration risk structure strongly associated with specific work processes, whereas fall-related accidents demonstrated a complex risk structure arising from the overlap of multiple work and spatial conditions. Furthermore, analysis of patterns related to safety facilities and personal protective equipment revealed that accidents may still occur despite the implementation of safety measures, indicating structural limitations in existing accident prevention systems. By extending conventional single-rule?centered ARM results to the level of accident scenarios through multi-rule combination, the proposed automated pattern derivation process provides an interpretable representation of recurring accident structures. This approach is expected to serve as a foundational analytical framework for identifying high-risk construction activities, establishing safety management priorities, and supporting intelligent construction safety management systems.