Scholarlink

<p>A database of 234 documented gravelly soil case histories from 17 earthquakes was analyzed to assess liquefaction potential using an interpretable machine learning framework. Data screening was performed before model development, and variables containing implicit outcome information were removed during preprocessing. The dataset was divided using a stratified training and testing approach, with 80% of records used for training and 20% reserved for testing. Model performance was evaluated using a receiver operating characteristic curve, and the area under the curve was approximately 0.92. Stability was checked using six independent data splits and four test set proportions, with area under the curve values ranging from 0.865 to 0.967. Feature contribution analysis showed that effective vertical stress and penetration resistance were the dominant predictors, with mean absolute contribution values of 1.99 and 1.61, respectively. Hydraulic boundary parameters were then examined as secondary conditioning factors. Capping layer thickness and unsaturated zone thickness showed nonlinear conditioning effects, with mean absolute contribution values of 0.36 and 0.22, respectively. Dependence analysis indicated a threshold near 4&#xa0;m for unsaturated zone thickness, beyond which predicted liquefaction probability decreased. Gravelly soil liquefaction is mainly controlled by stress and resistance variables, while hydraulic boundary parameters provide secondary conditioning effects.</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

Interpretable Machine Learning Assessment of Gravelly Soil Liquefaction with Hydraulic Boundary Conditioning and Interaction Effects

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Ameen A. Alshaba

摘要

A database of 234 documented gravelly soil case histories from 17 earthquakes was analyzed to assess liquefaction potential using an interpretable machine learning framework. Data screening was performed before model development, and variables containing implicit outcome information were removed during preprocessing. The dataset was divided using a stratified training and testing approach, with 80% of records used for training and 20% reserved for testing. Model performance was evaluated using a receiver operating characteristic curve, and the area under the curve was approximately 0.92. Stability was checked using six independent data splits and four test set proportions, with area under the curve values ranging from 0.865 to 0.967. Feature contribution analysis showed that effective vertical stress and penetration resistance were the dominant predictors, with mean absolute contribution values of 1.99 and 1.61, respectively. Hydraulic boundary parameters were then examined as secondary conditioning factors. Capping layer thickness and unsaturated zone thickness showed nonlinear conditioning effects, with mean absolute contribution values of 0.36 and 0.22, respectively. Dependence analysis indicated a threshold near 4 m for unsaturated zone thickness, beyond which predicted liquefaction probability decreased. Gravelly soil liquefaction is mainly controlled by stress and resistance variables, while hydraulic boundary parameters provide secondary conditioning effects.