抗压强度
超参数
随机森林
梯度升压
均方误差
Boosting(机器学习)
阿达布思
计算机科学
回归
决策树
集成学习
机器学习
过度拟合
数学
统计
人工智能
人工神经网络
支持向量机
材料科学
复合材料
作者
Woubishet Zewdu Taffese,Leonardo Espinosa-Leal
标识
DOI:10.1016/j.jobe.2023.106523
摘要
This work develops single multitarget regression models that predict compressive strength and non-steady-state chloride migration coefficients (Dnssm) of concrete simultaneously using machine learning algorithms. The data for this study are obtained from research projects and internationally published articles. Following data preprocessing, the compressive strength ranged from 21 to 80 MPa, while the Dnssm ranged from 1.57 to 31.30 × 10−12 m2/s. The algorithms used are five decision tree-based ensemble methods: bagging, random forest, AdaBoost, Gradient boosting, and XGBoost. In the development of the models, two scenarios are considered. Scenario 1 employs the default hyperparameter settings, while Scenario 2 employs hyperparameters chosen from among those identified through training single-target models. The performance evaluation results confirm that Gradient boosting is the best performing algorithm and Scenario 2 is the most appropriate modeling strategy for the considered dataset. It predicts compressive strength with (MAE = 6.683, MSE = 83.369, and RMSE = 9.131) and Dnssm with (MAE = 1.363, MSE = 3.712, and RMSE = 1.927). The potential of the developed multitarget model to design concrete with the intended strength and Dnssm is supported by its remarkable generalization ability. However, in order to ensure the model's versatility, it is necessary to improve it by incorporating comprehensive datasets that include a broad range of concrete properties.
科研通智能强力驱动
Strongly Powered by AbleSci AI