This paper presents a data-driven framework for inferring spatial interface dependencies in large-scale civil infrastructure design using empirically observed co-change as a proxy for coordination-related interactions. Using tender and as-built 3D models from an expressway project, object-level changes are identified and distance-based inter-category features are constructed to predict change occurrence. Among the evaluated classifiers, LightGBM achieved the best performance (average G-mean = 0.903). The results show that co-change is better explained by neighborhood-level spatial interactions than by single nearest-neighbor relationships. Distance cutoff analysis indicates that predictive performance saturates within limited spatial ranges, typically within about 30 m in this dataset. Furthermore, Individual Conditional Expectation (ICE) analysis reveals five distinct distance-dependent co-change patterns, while also identifying cases where distance reflects broader spatial heterogeneity rather than direct interface relationships. These findings suggest that spatial distance captures a substantial portion of co-change behavior and can support prioritization of interface checks and selection of search radii in infrastructure design coordination.
- Inferring Spatial Interface Dependencies from Co-change Records in Civil Infrastructure Design Models
- Ryoko ARASHIDA [嵐田, 涼子] (Corresponding Author)
- University of Tokyo, Institute of Engineering Innovation, Graduate School of Engineering
- Masahide Horita (Author)
- University of Tokyo, Department of Civil Engineering, Graduate School of Engineering
- Proceedings of the ... ISARC, pp.1666-1673
- 2413-5844
- 43rd International Symposium on Automation and Robotics in Construction (Singapore, Singapore, 2026/06/22–2026/06/26)
- 1666-1673
- Institute of Engineering Innovation, Graduate School of Engineering
- English
- Conference paper
- 2026