Scholarlink

<p>Geotechnical parameters, including the natural water content (Wn), liquid limit (WL), initial void ratio (e<sub>0</sub>), and dry density (Gd), were used to predict the soil compression index (Cc). Five Multiple Linear Regression (MLR) models were first developed to establish baseline relationships, which were subsequently enhanced using Artificial Neural Networks (ANN) to capture complex nonlinear interactions. The evaluation extended beyond standard metrics, such as the Root Mean Square Error (RMSE) and the coefficient of determination (R<sup>2</sup>), and also included in-depth statistical analyses. These include residual examination, multicollinearity verification via the Variance Inflation Factor (VIF), and sensitivity analysis. The results demonstrate that the Artificial Neural Networks model significantly outperforms Multiple Linear Regression, with the sensitivity analysis identifying e₀ as the primary predictor, while the other parameters showed a moderate influence on the prediction of compression index. The Variance Inflation Factor values remained below the critical thresholds, and residual analysis confirmed the absence of systematic bias, validating the friability, precision, and robustness of the Artificial Neural Networks approach for modeling compression index variability.</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

Robust Evaluation Framework for Compression Index Prediction Models

  • 期刊论文

  • 发表日期：

    2026年7月16日

查看全文
  • Hafida Berrich,
  • Souad Amal Bourokba,
  • Abdelkader Hachichi,
  • Mohamed Bousmaha

摘要

Geotechnical parameters, including the natural water content (Wn), liquid limit (WL), initial void ratio (e0), and dry density (Gd), were used to predict the soil compression index (Cc). Five Multiple Linear Regression (MLR) models were first developed to establish baseline relationships, which were subsequently enhanced using Artificial Neural Networks (ANN) to capture complex nonlinear interactions. The evaluation extended beyond standard metrics, such as the Root Mean Square Error (RMSE) and the coefficient of determination (R2), and also included in-depth statistical analyses. These include residual examination, multicollinearity verification via the Variance Inflation Factor (VIF), and sensitivity analysis. The results demonstrate that the Artificial Neural Networks model significantly outperforms Multiple Linear Regression, with the sensitivity analysis identifying e₀ as the primary predictor, while the other parameters showed a moderate influence on the prediction of compression index. The Variance Inflation Factor values remained below the critical thresholds, and residual analysis confirmed the absence of systematic bias, validating the friability, precision, and robustness of the Artificial Neural Networks approach for modeling compression index variability.