Scholarlink

<p>Safe design of a foundation is a major concern for structures, and recently machine learning (ML)-based reliability analysis has received considerable attention in academics. The study utilizes eXtreme Gradient Boosting Machine (XGBoost) and Multivariate Adaptive Regression Splines (MARS) to propose an interpretable, reliability-oriented predictive framework for estimating the reduction factor (RF) of eccentrically and obliquely loaded strip footings. Both models are trained and tested on a dataset of size 920 instances and 5 input parameters, collected from the literature. The performance of the models is evaluated using state-of-the-art performance indices. While XGBoost model has the edge with a slightly higher coefficient of determination (0.967), MARS has the edge with less training time and better interpretability with readymade equations for the prediction of output. The reliability indices are calculated for both models, and both models are concluded to be reliable methodologies for predicting the bearing capacity of eccentrically loaded strip footing using RF; however, XGBoost has a slight edge over MARS. The proposed framework enables rapid estimation of the RF and associated reliability index without repeated finite element analyses and experimental testing, thereby supporting practical geotechnical design and preliminary reliability assessment of eccentrically loaded strip foundations.</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 for reliability-oriented prediction of the reduction factor of eccentrically loaded strip footings using MARS and XGBoost

  • 期刊论文

  • 发表日期：

    2026年8月4日

查看全文
  • Manish Kumar

摘要

Safe design of a foundation is a major concern for structures, and recently machine learning (ML)-based reliability analysis has received considerable attention in academics. The study utilizes eXtreme Gradient Boosting Machine (XGBoost) and Multivariate Adaptive Regression Splines (MARS) to propose an interpretable, reliability-oriented predictive framework for estimating the reduction factor (RF) of eccentrically and obliquely loaded strip footings. Both models are trained and tested on a dataset of size 920 instances and 5 input parameters, collected from the literature. The performance of the models is evaluated using state-of-the-art performance indices. While XGBoost model has the edge with a slightly higher coefficient of determination (0.967), MARS has the edge with less training time and better interpretability with readymade equations for the prediction of output. The reliability indices are calculated for both models, and both models are concluded to be reliable methodologies for predicting the bearing capacity of eccentrically loaded strip footing using RF; however, XGBoost has a slight edge over MARS. The proposed framework enables rapid estimation of the RF and associated reliability index without repeated finite element analyses and experimental testing, thereby supporting practical geotechnical design and preliminary reliability assessment of eccentrically loaded strip foundations.