Interpretable Machine Learning in Construction Crane Accident Damage Estimation Using SHapely Additive exPlanations

Cranes in construction, due to their heavy nature, are a major cause of accidents, often leading to hazardous situations involving significant damage and fatalities. Utilizing 701 crane accident data from 1983 to 2017, this research aims to identify key factors influencing accident damage predictions across diverse data types, including numerical, categorical, and text-based features. An interpretable machine learning technique, namely, SHapley Additive exPlanation, was performed to identify the most influential features from machine learning-based prediction models. The analysis revealed that among the top nine features, the numerical feature “boom length” was highly influential, while the categorical feature indicating whether the crane’s outriggers were fully extended also ranked prominently. Additionally, more than 70% of the top 10 features were derived from the vector representations of text-based columns. The findings of this study contribute to enhancing crane safety practices by helping to reduce high-impact damage accidents and provide a foundational direction for applying advanced artificial intelligence-based prediction models focusing on textual natural language data.