Scholarlink

<p>We employed machine learning approaches and visualization interpretation methods to explore the influencing factors of the compressive strength of sea sand concrete to attain a better understanding of the inherent laws of concrete mix design. Four models, including random forest, CatBoost, XGBoost, and deep neural network, were trained. The experimental results demonstrate that the XGBoost model performs the best in predicting the strength of sea sand concrete. Its <i>R</i><sup>2</sup> value reached 0.9999, and evaluation indexes such as MAPE, RMSE, MAE, and MSE are superior to those of other models. The principal component analysis (PCA) was conducted to visually analyze the structure and distribution of the original feature data, and Pearson correlation coefficient analysis and Shapley additive explanation (SHAP) were utilized to explore the impact of input characteristics on the strength of sea sand concrete. SHAP analysis is more conducive to revealing the nonlinear effects of various characteristics on the model prediction results, especially that particle size of stone has significant impacts on the strength of sea sand concrete. In addition, experimental verification was carried out to confirm the accuracy of the optimized training model. These findings offer some insights for the future design and application of sea sand in high-performance marine and coastal infrastructure.</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

Visual Interpretation of Crucial Influencing Factors in Sea Sand Concrete Strength with Machine Learning Prediction

  • 期刊论文

  • 发表日期：

    2026年4月11日

查看全文
  • Naishu Zhu,
  • Fengnian Jin,
  • Zhongwen Ou,
  • Yinsuo Dai,
  • Yong Liu,
  • Zhi Zhang,
  • Linjian Ma,
  • Huguang He,
  • Hansong Zhang

摘要

We employed machine learning approaches and visualization interpretation methods to explore the influencing factors of the compressive strength of sea sand concrete to attain a better understanding of the inherent laws of concrete mix design. Four models, including random forest, CatBoost, XGBoost, and deep neural network, were trained. The experimental results demonstrate that the XGBoost model performs the best in predicting the strength of sea sand concrete. Its R2 value reached 0.9999, and evaluation indexes such as MAPE, RMSE, MAE, and MSE are superior to those of other models. The principal component analysis (PCA) was conducted to visually analyze the structure and distribution of the original feature data, and Pearson correlation coefficient analysis and Shapley additive explanation (SHAP) were utilized to explore the impact of input characteristics on the strength of sea sand concrete. SHAP analysis is more conducive to revealing the nonlinear effects of various characteristics on the model prediction results, especially that particle size of stone has significant impacts on the strength of sea sand concrete. In addition, experimental verification was carried out to confirm the accuracy of the optimized training model. These findings offer some insights for the future design and application of sea sand in high-performance marine and coastal infrastructure.