ANN 및 데이터 샘플링 기반 항만 터미널 작업 현장 사고 심각도 요인 분석

초록

In port terminals, workers and numerous heavy equipment operate in close proximity, and severity can be high in the event of accidents. Therefore, appropriate attention and safety measures are required to prevent serious accidents. This study employed machine learning methodologies, namely artificial neural network (ANN), to develop models for predicting port accident severity. Additionally, oversampling and clustering techniques were applied to augment data for all features of port accident data and to separate accident risk situations based on features. Consequently, a port accident severity model was developed using heterogeneous accident data clustered by features, and its performance was compared and analyzed against a severity model built on non-clustered accident data. Research findings indicate that accident models utilizing clustered data reflecting the heterogeneity of port accident data show better performance than accident models utilizing non-clustered data. Finally, to explain the machine learning model, the ANN model was used to perform explainable artificial intelligence (XAI) analysis and identify key variables influencing severe accidents in ports. This study introduces a high-performance and interpretable analytics model for assessing port accident severity, offering valuable insights for informed decision-making by experts. The findings of this study can play a crucial role in reducing accident severity within port terminals, promoting safety, and contributing to reducing economic losses caused by accidents in ports.

키워드