Scholarlink

<p>Three popular regression artificial intelligence (AI) methods such as adaptive boosting (AGB), random forest (RF), and multi-layer perceptron (MLP) were employed in this study for forecasting the compressive strength of lightweight foamed concrete (LFC). These models have been trained and tested based on 91 data samples collected from the open literatures. Four potential engineering features, including cement, over dry density, water binder ratio, and foam volume, were selected as input variable. The results indicate that the RF model demonstrates strong predictive capability for estimating the compressive strength of LFC, as evidenced by its high performance across key evaluation metrics (<i>R</i><sup>2</sup> = 0.963 for training, 0.943 for testing, and 0.720 following 10-fold cross-validation). In addition, the RF model has the low uncertainty value with U<sub>95</sub> = 5.519 indicating that high predicative reliability. Sensitivities analysis via Shapely Additive Explanations (SHAP) and Partial Dependence Plots proved that oven dry density is most important input variables significantly influenced on the compressive strength of LFC. Finally, an effective web application has been built and validated for rapidly predicting the LFC’s compressive strength in the material design.</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

Forecasting Compressive Strength of Lightweight Foamed Concrete Using Regression AI Technique and Parametric Investigation

  • 期刊论文

  • 发表日期：

    2026年1月30日

查看全文
  • Lieu B. Nguyen

摘要

Three popular regression artificial intelligence (AI) methods such as adaptive boosting (AGB), random forest (RF), and multi-layer perceptron (MLP) were employed in this study for forecasting the compressive strength of lightweight foamed concrete (LFC). These models have been trained and tested based on 91 data samples collected from the open literatures. Four potential engineering features, including cement, over dry density, water binder ratio, and foam volume, were selected as input variable. The results indicate that the RF model demonstrates strong predictive capability for estimating the compressive strength of LFC, as evidenced by its high performance across key evaluation metrics (R2 = 0.963 for training, 0.943 for testing, and 0.720 following 10-fold cross-validation). In addition, the RF model has the low uncertainty value with U95 = 5.519 indicating that high predicative reliability. Sensitivities analysis via Shapely Additive Explanations (SHAP) and Partial Dependence Plots proved that oven dry density is most important input variables significantly influenced on the compressive strength of LFC. Finally, an effective web application has been built and validated for rapidly predicting the LFC’s compressive strength in the material design.