RAGN-R: A multi-subject ensemble machine-learning method for estimating mechanical properties of advanced structural materials

• Published in: Computers & structures Vol. 308; p. 107657
• Main Authors: Kazemi, F., Ӧzyüksel Çiftçioğlu, A., Shafighfard, T., Asgarkhani, N., Jankowski, R.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2025
• Subjects: Data processing; Ensemble machine-learning model; Fiber-reinforced concrete; Fiber-reinforced concrete beam; Multi-subject machine-learning model; Preplaced aggregate concrete; Data processing; Multi-subject machine-learning model; Preplaced aggregate concrete; Fiber-reinforced concrete; Fiber-reinforced concrete beam; Ensemble machine-learning model
• ISSN: 0045-7949
• DOI: 10.1016/j.compstruc.2025.107657

•Multi-subject ensemble ML method proposed to estimate mechanical properties of materials.•Automated optimizing and utilizing hyperparameters exhibited superior ML performance.•Proposed RAGN-R model is hyperparameter free to estimate mechanical properties.•Novel RAGN-R model showed superior accuracies compared to other well-known ML algorithms. The utilization of advanced structural materials, such as preplaced aggregate concrete (PAC), fiber-reinforced concrete (FRC), and FRC beams has revolutionized the field of civil engineering. These materials exhibit enhanced mechanical properties compared to traditional construction materials, offering engineers unprecedented opportunities to optimize the design, construction, and performance of structures and infrastructures. This formal description elucidates the inherent mechanical properties of PAC, FRC, and FRC beams, explores their diverse applications in civil engineering projects. This research aims to propose a surrogate multi-subject ensemble machine-learning (ML) method (named RAGN-R) for estimating mechanical properties of aforementioned advanced materials. The proposed learning approach, RAGN-R, integrates Random forest, Adaptive boosting, and GradieNt boosting techniques, employing a Ridge regression framework for stacking the ensemble. For this purpose, three experimental dataset have been prepared to determine the capability of RAGN-R and the results of the study have been compared with six well-known ML models. It is noteworthy that the proposed RAGN-R has the ability of self-optimizing the hyperparameters, which facilitate the adoptability of the model with engineering problems. Moreover, three datasets have been investigated to show the ability of the RAGN-R for diverse problems. Different performance evaluation metrics have been conducted to present results and compare ML models, which confirms the highest performance of RAGN-R (i.e., 97.7% accuracy) in handling complex relationships and improving overall prediction accuracy.