Scholarlink

<p>This study presents an integrated experimental–predictive framework to estimate rainfall-induced deformation in unsaturated clay slopes using flume tests and lightweight, interpretable machine learning models. Flume-scale experiments were conducted on sand–clay mixtures with clay contents ranging from 5 to 20%, subjected to a controlled rainfall intensity of 100&#xa0;mm/h. Crest settlement and volumetric water content were monitored over time to capture hydromechanical responses. A threshold clay content between 15% and 17.5% was identified, beyond which infiltration accelerated and deformation intensified. Settlement data were used to train and evaluate seven regression models: support vector regression, gradient boosting, random forest, decision tree, k-nearest neighbours, and polynomial regression. Models were assessed under default and hyperparameter-tuned conditions. Gradient boosting achieved the highest baseline accuracy, while tuning improved support vector regression performance to a coefficient of determination of 0.982 and a root mean square error of 1.12&#xa0;mm. Sensitivity analysis showed time as the most influential predictor variable. The findings demonstrate a practical, sensor-integrated framework that can support early-stage stability assessments and, with further development, may inform low-cost monitoring or early-warning applications in rainfall-prone geotechnical infrastructure.</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

Data-Driven Prediction of Rainfall-Triggered Slope Movements Using Flume Tests and Interpretable Machine Learning Models

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • Meghdad Bagheri

摘要

This study presents an integrated experimental–predictive framework to estimate rainfall-induced deformation in unsaturated clay slopes using flume tests and lightweight, interpretable machine learning models. Flume-scale experiments were conducted on sand–clay mixtures with clay contents ranging from 5 to 20%, subjected to a controlled rainfall intensity of 100 mm/h. Crest settlement and volumetric water content were monitored over time to capture hydromechanical responses. A threshold clay content between 15% and 17.5% was identified, beyond which infiltration accelerated and deformation intensified. Settlement data were used to train and evaluate seven regression models: support vector regression, gradient boosting, random forest, decision tree, k-nearest neighbours, and polynomial regression. Models were assessed under default and hyperparameter-tuned conditions. Gradient boosting achieved the highest baseline accuracy, while tuning improved support vector regression performance to a coefficient of determination of 0.982 and a root mean square error of 1.12 mm. Sensitivity analysis showed time as the most influential predictor variable. The findings demonstrate a practical, sensor-integrated framework that can support early-stage stability assessments and, with further development, may inform low-cost monitoring or early-warning applications in rainfall-prone geotechnical infrastructure.