The International Association for Automation and Robotics in Construction
Publications / CCC 2025 - Zadar, Croatia
AUTOMATED CLASSIFICATION OF CONSTRUCTION WORKERS' WALKING-RELATED ACCIDENTS IN ACCIDENT CASE DATABASE
Construction workers' walking-related accidents, which refer to incidents occurring while construction workers transition between work areas rather than during task execution, account for 31% of construction site incidents and often result from workers' lack of awareness of site hazards. Despite their high frequency, existing accident classification methods primarily focus on site characteristics, accident types, or worker attributes, making it difficult to analyze these accidents sufficiently. Since these methods are designed to assess incidents that occur during task execution, they often fail to capture the distinct nature of walking-related accidents. This study proposes a structured methodology for the automated classification of walking-related accidents using Natural Language Processing (NLP) techniques. We define classification criteria tailored to walking-related accidents on construction sites and apply a supervised learning approach using Bidirectional Encoder Representations from Transformers (BERT) fine-tuning and machine learning classifiers. This method enables precise identification and categorization of walking-related accidents from textual accident data. To validate this approach, we conducted a case study using the Korean construction accident dataset. The study confirmed that the proposed classifier model achieved a high precision score of 0.93. Through 1,360 labeled cases, we identified 7,720 walking-related accident cases from 23,469 reports, demonstrating the effectiveness of our classification framework in extracting walking-related accident case data. This study contributes to improving construction site safety management by shifting the focus from conventional task-centered accident classification to the specific classification of walking-related accidents. Furthermore, it enhances the automation of the classification process, enabling more efficient handling of large-scale accident case data and addressing the limitations of expert-dependent classification methods. The classified walking-related accident data supports worker-friendly safety education, strengthens risk assessment for walking hazards, and provides valuable insights for developing proactive accident prevention strategies.