Scholarlink

<p>This study highlights the importance of consistently assessing soil and subsoil liquefaction susceptibility in construction projects through comprehensive evaluation and regular site investigations. Despite the abundance of geotechnical reports, many remain underutilized post-project completion, while the literature contains numerous site-specific case studies. To address this gap, this study suggests integrating site-specific reports tailored to both the site and the region, aiming to extract valuable insights into liquefaction potential. Utilizing ensemble-based machine learning techniques, two databases of soil liquefaction were analyzed to predict the probability of liquefaction failure (PLF). The results reveal the superior accuracy of the Gradient Boosting Regressor (GBR) model, achieving an almost ideal accuracy in predicting PLF. Despite its generalization across diverse geographical patterns, the GBR model encounters limitations with insufficient experimental data within specific parameter ranges. To address this, a graphical user interface (GUI) was developed by leveraging data from previous liquefaction records to predict PLF. The GUI, which has been included as supplementary material, proved to be a useful tool for liquefaction risk assessment, and its predictive capabilities make it invaluable for both practical applications and educational purposes. This empowers engineers, urban planners, and decision-makers to make informed decisions and implement proactive measures for disaster mitigation and infrastructure development resilience.</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

Revealing the nature of soil liquefaction using machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年1月21日

查看全文
  • Sufyan Ghani,
  • Ishwor Thapa,
  • Sunita Kumari,
  • Antonio Gomes Correia,
  • Panagiotis G. Asteris

摘要

This study highlights the importance of consistently assessing soil and subsoil liquefaction susceptibility in construction projects through comprehensive evaluation and regular site investigations. Despite the abundance of geotechnical reports, many remain underutilized post-project completion, while the literature contains numerous site-specific case studies. To address this gap, this study suggests integrating site-specific reports tailored to both the site and the region, aiming to extract valuable insights into liquefaction potential. Utilizing ensemble-based machine learning techniques, two databases of soil liquefaction were analyzed to predict the probability of liquefaction failure (PLF). The results reveal the superior accuracy of the Gradient Boosting Regressor (GBR) model, achieving an almost ideal accuracy in predicting PLF. Despite its generalization across diverse geographical patterns, the GBR model encounters limitations with insufficient experimental data within specific parameter ranges. To address this, a graphical user interface (GUI) was developed by leveraging data from previous liquefaction records to predict PLF. The GUI, which has been included as supplementary material, proved to be a useful tool for liquefaction risk assessment, and its predictive capabilities make it invaluable for both practical applications and educational purposes. This empowers engineers, urban planners, and decision-makers to make informed decisions and implement proactive measures for disaster mitigation and infrastructure development resilience.