Scholarlink

<p>This study investigates the predictive modeling of unconfined compressive strength (UCS) in soil-fly ash systems enhanced with multi-walled carbon nanotubes (MWCNTs). Advanced regression techniques, including Least Absolute Shrinkage and Selection Operator (LASSO), Ridge, and Elastic net, were employed to develop robust models for UCS prediction. The experimental design incorporated up to 50% fly ash replacement and MWCNT doping levels of 0%, 0.01%, and 0.001%, enabling a comprehensive analysis of their combined effects. Results indicate that Elastic net achieved the best performance during training, with the highest R<sup>2</sup> value (0.813) and lowest error metrics, but LASSO outperformed all models in testing, demonstrating superior generalizability with a testing R<sup>2</sup> of 0.781 and the lowest root mean square error (RMSE) (0.211). LASSO’s capacity for automatic feature selection and emphasis on critical predictors (Cement: 0.297; Sodium hexametaphosphate (SHMP): 0.053; Days: 0.013) highlights its suitability for real-world applications requiring simplicity and interpretability. Comparative analysis of feature importance revealed that Cement (%) consistently emerged as the dominant predictor across all models, with secondary contributions from SHMP (%) and optimum moisture content (OMC). LASSO proved ideal for scenarios prioritizing key variable identification, while Ridge excelled in retaining all predictors with balanced shrinkage, and Elastic net provided a hybrid approach for correlated features. These findings establish LASSO as the most robust model for practical deployment, balancing accuracy, generalizability, and interpretability. The robustness of LASSO model is validated by the Taylor diagrams.</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

Machine learning-based hybrid regularization techniques for predicting unconfined compressive strength of soil reinforced with multiple additives

  • 期刊论文

  • 发表日期：

    2025年3月20日

查看全文
  • Anish Kumar,
  • Sanjeev Sinha

摘要

This study investigates the predictive modeling of unconfined compressive strength (UCS) in soil-fly ash systems enhanced with multi-walled carbon nanotubes (MWCNTs). Advanced regression techniques, including Least Absolute Shrinkage and Selection Operator (LASSO), Ridge, and Elastic net, were employed to develop robust models for UCS prediction. The experimental design incorporated up to 50% fly ash replacement and MWCNT doping levels of 0%, 0.01%, and 0.001%, enabling a comprehensive analysis of their combined effects. Results indicate that Elastic net achieved the best performance during training, with the highest R2 value (0.813) and lowest error metrics, but LASSO outperformed all models in testing, demonstrating superior generalizability with a testing R2 of 0.781 and the lowest root mean square error (RMSE) (0.211). LASSO’s capacity for automatic feature selection and emphasis on critical predictors (Cement: 0.297; Sodium hexametaphosphate (SHMP): 0.053; Days: 0.013) highlights its suitability for real-world applications requiring simplicity and interpretability. Comparative analysis of feature importance revealed that Cement (%) consistently emerged as the dominant predictor across all models, with secondary contributions from SHMP (%) and optimum moisture content (OMC). LASSO proved ideal for scenarios prioritizing key variable identification, while Ridge excelled in retaining all predictors with balanced shrinkage, and Elastic net provided a hybrid approach for correlated features. These findings establish LASSO as the most robust model for practical deployment, balancing accuracy, generalizability, and interpretability. The robustness of LASSO model is validated by the Taylor diagrams.