From Research Trend to Performance Prediction: Metaheuristic-Driven Machine Learning Optimization for Cement Pastes...

Area of Science:

• Materials Science
• Civil Engineering
• Computational Science

Background:

• Phase change materials (PCMs) enhance thermal energy storage in cementitious composites.
• Predicting the coupled thermal and mechanical properties of these advanced materials is challenging.
• Research trends in bio-based PCMs for cement pastes require systematic exploration.

Purpose of the Study:

• To identify current research hotspots in cement pastes with bio-based PCMs using bibliometric analysis.
• To develop and optimize machine learning models for predicting thermal conductivity (Tc), latent heat capacity (LH), and compressive strength (CS).
• To establish a data-driven pipeline for evaluating the performance of cement pastes containing bio-based PCMs.

Main Methods:

• Bibliometric analysis of 5928 articles using CiteSpace to map research trends.
• Compilation of a dataset with 100 experimental samples and nine input variables.
• Optimization of four machine learning algorithms (SVR, RF, XGBoost, CatBoost) using five metaheuristic algorithms (GA, PSO, WOA, GWO, FFA).

Main Results:

• The CatBoost-WOA hybrid model demonstrated superior predictive performance.
• Achieved high R-squared values (0.927-0.955) and low RMSEs (0.0057-2.91) for Tc, LH, and CS.
• SVR-GWO and XGBoost-WOA models also exhibited strong generalization capabilities.

Conclusions:

• The integrated bibliometric and machine learning approach effectively maps research trends and predicts material performance.
• Optimized hybrid models offer a reliable method for assessing cement pastes with bio-based PCMs.
• The developed modeling pipeline facilitates the design and application of sustainable building materials.