Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable AI models for predicting liquefaction-induced lateral spreading

Cheng-Hsi Hsiao, Krishna Kumar, Ellen Rathje

TL;DR

This study tackles the interpretability challenge of ML-based liquefaction-induced lateral spreading prediction by applying SHAP explanations to an XGBoost model trained on the 2011 Christchurch dataset. The approach reveals how features such as distance to river, groundwater depth, and CPT-derived soil properties drive predictions, with local SHAP analyses validating alignment to domain knowledge in many cases and exposing counterintuitive PGA relationships. Global explanations identify PGA as the most influential feature yet highlight non-monotonic, interaction-driven patterns that challenge simple physical intuition. CPT features provide valuable explanatory context but do not always improve predictive accuracy, guiding a move toward simpler, more explainable models for reliable hazard assessment in geotechnical engineering.

Abstract

Earthquake-induced liquefaction can cause substantial lateral spreading, posing threats to infrastructure. Machine learning (ML) can improve lateral spreading prediction models by capturing complex soil characteristics and site conditions. However, the "black box" nature of ML models can hinder their adoption in critical decision-making. This study addresses this limitation by using SHapley Additive exPlanations (SHAP) to interpret an eXtreme Gradient Boosting (XGB) model for lateral spreading prediction, trained on data from the 2011 Christchurch Earthquake. SHAP analysis reveals the factors driving the model's predictions, enhancing transparency and allowing for comparison with established engineering knowledge. The results demonstrate that the XGB model successfully identifies the importance of soil characteristics derived from Cone Penetration Test (CPT) data in predicting lateral spreading, validating its alignment with domain understanding. This work highlights the value of explainable machine learning for reliable and informed decision-making in geotechnical engineering and hazard assessment.

Explainable AI models for predicting liquefaction-induced lateral spreading

TL;DR

This study tackles the interpretability challenge of ML-based liquefaction-induced lateral spreading prediction by applying SHAP explanations to an XGBoost model trained on the 2011 Christchurch dataset. The approach reveals how features such as distance to river, groundwater depth, and CPT-derived soil properties drive predictions, with local SHAP analyses validating alignment to domain knowledge in many cases and exposing counterintuitive PGA relationships. Global explanations identify PGA as the most influential feature yet highlight non-monotonic, interaction-driven patterns that challenge simple physical intuition. CPT features provide valuable explanatory context but do not always improve predictive accuracy, guiding a move toward simpler, more explainable models for reliable hazard assessment in geotechnical engineering.

Abstract

Earthquake-induced liquefaction can cause substantial lateral spreading, posing threats to infrastructure. Machine learning (ML) can improve lateral spreading prediction models by capturing complex soil characteristics and site conditions. However, the "black box" nature of ML models can hinder their adoption in critical decision-making. This study addresses this limitation by using SHapley Additive exPlanations (SHAP) to interpret an eXtreme Gradient Boosting (XGB) model for lateral spreading prediction, trained on data from the 2011 Christchurch Earthquake. SHAP analysis reveals the factors driving the model's predictions, enhancing transparency and allowing for comparison with established engineering knowledge. The results demonstrate that the XGB model successfully identifies the importance of soil characteristics derived from Cone Penetration Test (CPT) data in predicting lateral spreading, validating its alignment with domain understanding. This work highlights the value of explainable machine learning for reliable and informed decision-making in geotechnical engineering and hazard assessment.
Paper Structure (13 sections, 3 equations, 4 figures, 2 tables)

This paper contains 13 sections, 3 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. eXtreme Gradient Boosting
  4. SHAP
  5. Model Development
  6. Results
  7. Local explanations
  8. True positives and true negatives
  9. False positive and false negative
  10. Global explanations
  11. Incorrect learning of PGA
  12. Effect of CPT on Lateral Spreading Predictions
  13. Conclusion

Figures (4)

  • Figure 3: Developing sequence of an XGB model
  • Figure 4: An example of SHAP calculation for factors influencing lateral spreading.
  • Figure 11: Pairplots of CPT features with lateral spreading observations.
  • Figure 16: SHAP values for medians of $I_c$ in Model C