All Issue

2025 Vol.26, Issue 12 Preview Page
Explainable AI-Based Analysis of Sinkhole Risk Factors Using SHAP and Identification of Density-Based Management Zones

SHAP 기반 설명가능한 AI를 활용한 지반함몰 위험도 영향요인 분석 및 밀집도 기반 관리구간 제안

Sungyeol Lee1
,
Jaemo Kang2
,
Myeongsik Kong3
,
Wonchul Jeong4
,
Jinyoung Kim5*
이 성열1
,
강 재모2
,
공 명식3
,
정 원철4
,
김 진영5*
1Postdoctoral Researcher, Department of Geotechnical Engineering Research, Korea Institute of Civil Engineering and Building Technology
2Research Fellow, Department of Geotechnical Engineering Research, Korea Institute of Civil Engineering and Building Technology, Senior Researcher
3Research Specialist, Department of Geotechnical Engineering Research, Korea Institute of Civil Engineering and Building Technology, Senior Researcher
4Student Researcher, Department of Geotechnical Engineering Research, Korea Institute of Civil Engineering and Building Technology, Senior Researcher
5Senior Researcher, Department of Geotechnical Engineering Research, Korea Institute of Civil Engineering and Building Technology
*Corresponding Author
1 December 2025. pp. 13-19
PDF XML Citation
Abstract

Ground subsidence is an emerging urban safety issue increasingly linked to aging underground infrastructure, excavation activities, and variability in environmental conditions. Despite recent advances in machine-learning–based risk prediction, the influence of spatial density of underground facilities has not been clearly quantified. This study aims to evaluate whether underground facility density can serve as a key factor for decision-making in ground-subsidence risk management. Gradient boosting–based classification models were developed using multi-source data, including facility attributes, excavation history, and environmental variables. A density index was generated for each analysis segment, and its relationship with predicted risk levels was examined using partial dependence analysis and threshold-based sensitivity evaluation. The results show that higher density values exhibit a clear positive relationship with high-risk predictions, and the most rapid increase in predicted risk occurred within a specific density threshold range. This indicates that facility density acts not only as a descriptive attribute but also as an effective decision-support criterion for prioritizing inspection and monitoring zones. The findings suggest that incorporating the density index into risk prediction and management strategies can improve the allocation of limited maintenance resources and support proactive ground-subsidence prevention.

Keywords
Ground Subsidence
SHAP
AI
Density
Risk management

도심지를 중심으로 상수, 하수, 전기, 통신, 난방, 가스 등 다양한 지하매설물이 집중적으로 매설되어 있으며, 매설관의 노후화와 반복적인 굴착 공사로 인하여 지반함몰 사고가 발생하고 있다. 지반함몰 사고는 도심지 중심에서 발생하기 때문에 사고 발생 시 인력 및 재산 피해를 야기하여 사회적 문제로 대두되고 있다. 따라서 사고 발생 가능성을 사전에 예측하고 관리할 수 있는 정량적 의사결정 체계가 필요하다. 본 연구는 기존 개발된 XGB(XGBoost) 기반 지반함몰 위험도 예측 모델에 설명가능한 인공지능인 SHAP을 적용하여, 지반함몰 위험도 예측 결과에 영향을 미치는 주요 인자를 분석하였다. 그 결과, 지하매설물의 밀집도와 활용년수가 위험도 분류에 가장 큰 영향을 미치는 인자로 확인되었으며, SHAP(SHapley Additive Explanations) 분석을 통해 밀집도가 증가할수록 지반함몰 위험도가 높아지는 방향으로 기여하는 경향을 확인하였다. 또한, 밀집도가 높은 구간에서 고위험 등급으로 분류될 확률이 증가하며, 밀집도가 0.05와 0.10을 기준으로 위험도 기여 방향의 변화가 발생하였다. 이를 통해 밀집도를 활용하여 지반함몰 우선관리구간을 정량적으로 구분할 수 있는 임계 기준값으로 해석 가능하다. 본 연구는 지반함몰 위험 예측 과정을 인자에 따른 메커니즘으로 가시화함으로써, 밀집도를 활용한 지반함몰 예방 및 유지관리 전략 수립에 활용 가능할 것으로 기대된다.

키워드
지반함몰
SHAP
AI
밀집도
위험관리
References
1

D. Takeuchi, W. Fukatani, T. Miyamoto and T. Yokota. (2017), Using decision tree analysis to extract factors affecting road subsidence, Journal of the Japan sewage works association, Vol. 54, No. 657, pp. 124~133.

2

Jin, Y. S. (2018), The analysis on correlation of precipitation and risk factors to the soil subsidence, Doctoral dissertation, Chonnam National University. (In Korean)

3

Kim, J. Y., Kang, J. M., Choi, C. H. and Park, D. H. (2017), Correlation Analysis of Sewer Integrity and Ground Subsidence, Journal of the Korean Geo-Environmental Society, Vol. 18, No. 6, pp. 31~37. (In Korean)

4

Kuwano, R., Horii, T., Kohashi, H. and Yamauchi, K. (2006), Defects of sewer pipes causing cave-in’s in the road, In Proceedings of the 5th International Symposium on New Technologies for Urban Safety of Mega Cities in Asia, Phuket, Thailand, pp. 16~17.

5

Lee, K., Nam, J., Park, J. and Hong, G. (2022a), Numerical Analysis of Factors Influencing the Ground Surface Settlement above a Cavity, Materials, Vol. 15m, No. 23, 8301.

10.3390/ma1523830136499798PMC9738637
6

Lee, S. Y., Kang, J. M. and Kim, J. Y. (2022b), Development of Machine Learning Model to predict the ground subsidence risk grade according to the Characteristics of underground facility, Journal of the Korean Geo-Environmental Society, Vol. 23, No. 4, pp. 5~10. (In Korean)

7

Lee, S. Y., Kang, J. M. and Kim, J. Y. (2023), Prediction modeling of ground subsidence risk based on machine learning using the attribute information of underground utilities in urban areas in Korea, Applied Sciences, Vol. 13, No. 9, 5566.

10.3390/app13095566
8

Lee, S. Y., Kang, J. M., Kim, J. Y. and Kong M. S. (2024), Development of a Regression Model for Predicting Ground Subsidence Risk Using Machine Learning, Journal of the Korean Society of Hazard Mitigation, Vol.24, No.6, pp. 329~336. (In Korean)

10.9798/KOSHAM.2024.24.6.329
9

Lee, S. Y., Kim, J. Y., Kang, J. M. and Baek, W. J. (2022c), Comparison of Machine Learning Models to Predict the Occurrence of Ground Subsidence According to the Characteristics of Sewer, Journal of Korean Geo-Environmental Society, Vol. 23, Issue. 4, pp. 5~10. (In Korean)

10

Li, Z. (2022), Extracting spatial effects from machine learning model using local interpretation method: An example of SHAP and XGBoost. Computers, Environment and Urban Systems, 96, 101845.

10.1016/j.compenvurbsys.2022.101845
11

Liu, M., Ning, Y., Yuan, H., Ong, M. E. H. and Liu, N. (2022), Balanced background and explanation data are needed in explaining deep learning models with SHAP: An empirical study on clinical decision making, arXiv preprint arXiv:2206.04050.

12

Moon, S. E., Jang, S. B., Lee, J. H. and Lee, J. S. (2016), Machine learning and deep learning technology trends, Journal of the Korean Society of Telecommunications (Information and Communication), Vol. 33, No. 10, pp. 49~56. (In Korean)

13

Mosca, E., Szigeti, F., Tragianni, S., Gallagher, D. and Groh, G. (2022), SHAP-based explanation methods: a review for NLP interpretability. In Proceedings of the 29th international conference on computational linguistics, pp. 4593~4603.

14

Mukunoki, T., Kuwano, N., Otani, J. and Kuwano, R. (2009), Visualization of three dimensional failure in sand due to water inflow and soil drainage from defected underground pipe using X-ray CT, Soils and Foundations, Vol. 49, No. 6, pp. 959~968.

10.3208/sandf.49.959
15

Seoul Seokchon-dong Cavity Cause Investigation Committee, (2014), Cause Analysis of Cavity at Seokchon Underground Roadway and Road Cavity.

16

Shapley, L. S. (1953), A value for n-person games.

10.1515/9781400881970-018
17

Van den Broeck, G., Lykov, A., Schleich, M. and Suciu, D. (2022), On the tractability of SHAP explanations, Journal of Artificial Intelligence Research, 74, pp. 851-886.

10.1613/jair.1.13283
18

You, S. K., Kim, J. B., Han, J. G., Hong, G. G., Yun, J. M. and Lee, K. I. (2017), A study on simulation of cavity and relaxation zone using finite element method. Journal of the Korean Geosynthetics Society, Vol. 16, No. 4, pp. 67~74. (In Korean)

Information
  • Publisher :Korean Geo-Environmental Society
  • Publisher(Ko) :한국지반환경공학회
  • Journal Title :Journal of the Korean Geo-Environmental Society
  • Journal Title(Ko) :한국지반환경공학회 논문집
  • Volume : 26
  • No :12
  • Pages :13-19
  • Received Date : 2025-11-11
  • Revised Date : 2025-11-12
  • Accepted Date : 2025-11-20
  • DOI :https://doi.org/10.14481/jkges.2025.26.12.13
Journal Informaiton Journal of the Korean Geo-Environmental Society Journal of the Korean Geo-Environmental Society
Online Submission
  • NRF
  • KOFST
  • KISTI Current Status
  • KISTI Cited-by
  • orcid
Journal Informaiton Journal Informaiton - close