Scholarlink

<p>The accurate prediction of liquefaction events is crucial for ensuring the safety and stability of structures during seismic events. Machine learning (ML) techniques such as boosting have been shown to be very effective in predicting the output. However, as the model’s complexity increases, the interpretation of the model’s prediction becomes more challenging. Thus, in order to achieve transparency and trust in ML models, explainable machine learning (EML) techniques can be applied. This study compares robust boosting classification algorithms such as AdaBoost, CatBoost, XGBoost, and LGBM. The embedded feature selection method and principal component analysis are used, and the performance of the ML models is evaluated and compared. Finally, two EML techniques, Shapley additive explanations (SHAP) and local interpretable model agnostic explanations (LIME), are used to interpret the predictions of the ML model. The results of the study suggest that developed ML models could reliably predict soil liquefaction, with the XGBoost model outperforming other models in terms of overall ranking. According to SHAP and LIME results, EML techniques perform considerably well in local and global explanations of the XGBoost model and are capable of addressing the gap between traditional knowledge in the field of liquefaction and ML methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction and interpretation of liquefaction occurrences using explainable machine learning models

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • Kishan Kumar,
  • Pijush Samui,
  • S. S. Choudhary

摘要

The accurate prediction of liquefaction events is crucial for ensuring the safety and stability of structures during seismic events. Machine learning (ML) techniques such as boosting have been shown to be very effective in predicting the output. However, as the model’s complexity increases, the interpretation of the model’s prediction becomes more challenging. Thus, in order to achieve transparency and trust in ML models, explainable machine learning (EML) techniques can be applied. This study compares robust boosting classification algorithms such as AdaBoost, CatBoost, XGBoost, and LGBM. The embedded feature selection method and principal component analysis are used, and the performance of the ML models is evaluated and compared. Finally, two EML techniques, Shapley additive explanations (SHAP) and local interpretable model agnostic explanations (LIME), are used to interpret the predictions of the ML model. The results of the study suggest that developed ML models could reliably predict soil liquefaction, with the XGBoost model outperforming other models in terms of overall ranking. According to SHAP and LIME results, EML techniques perform considerably well in local and global explanations of the XGBoost model and are capable of addressing the gap between traditional knowledge in the field of liquefaction and ML methods.