Scholarlink

<p>This study developed a hybrid machine learning framework for predicting the factor of safety against liquefaction using Gradient Boosting Machines optimized with Differential Evolution, Particle Swarm Optimization, and the Sparrow Search Algorithm. A site-specific dataset of 240 Standard Penetration Test records from the Barauni Refinery site in Bihar, India, was used, incorporating eight geotechnical and seismic input variables. The models were trained and evaluated using the coefficient of determination, root mean square error, and mean absolute error, along with uncertainty analysis to quantify predictive variability. External validation indices, including k, k′, Rₒ², were employed to assess generalization performance. Additional diagnostic tools such as regression error characteristic curves etc. were used to evaluate accuracy, dispersion, and correlation structure. Among the hybrid formulations, the Gradient Boosting–Differential Evolution model achieved the highest predictive performance and exhibited stable convergence under varying parameter conditions. A probabilistic assessment using the First Order Second Moment method further quantified the influence of input variability on failure probability and demonstrated consistency with model-based estimates. The results confirmed that the proposed hybrid Gradient Boosting framework captured the nonlinear behaviour of liquefiable soils and provided a reliable and scalable predictive tool for seismic hazard evaluation in data-limited geotechnical settings.</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 and Probabilistic Liquefaction Assessment Using Metaheuristically Tuned Gradient Boosting Frameworks

  • 期刊论文

  • 发表日期：

    2025年12月5日

查看全文
  • Chanchal Kumari,
  • Amit Kumar Verma

摘要

This study developed a hybrid machine learning framework for predicting the factor of safety against liquefaction using Gradient Boosting Machines optimized with Differential Evolution, Particle Swarm Optimization, and the Sparrow Search Algorithm. A site-specific dataset of 240 Standard Penetration Test records from the Barauni Refinery site in Bihar, India, was used, incorporating eight geotechnical and seismic input variables. The models were trained and evaluated using the coefficient of determination, root mean square error, and mean absolute error, along with uncertainty analysis to quantify predictive variability. External validation indices, including k, k′, Rₒ², were employed to assess generalization performance. Additional diagnostic tools such as regression error characteristic curves etc. were used to evaluate accuracy, dispersion, and correlation structure. Among the hybrid formulations, the Gradient Boosting–Differential Evolution model achieved the highest predictive performance and exhibited stable convergence under varying parameter conditions. A probabilistic assessment using the First Order Second Moment method further quantified the influence of input variability on failure probability and demonstrated consistency with model-based estimates. The results confirmed that the proposed hybrid Gradient Boosting framework captured the nonlinear behaviour of liquefiable soils and provided a reliable and scalable predictive tool for seismic hazard evaluation in data-limited geotechnical settings.