<p>Compressive strength (CS) is a key property of concrete mix, but the determination of CS requires costly and time intensive experimental procedures. Leveraging machine learning (ML) techniques for CS prediction can enhance accuracy and reliability while reducing the extensive need for laboratory testing.This study utilizes high performance concrete mix data set and employs the Extreme Gradient Boosting (XGBoost) ML model coupled with six metaheuristic optimization techniques such as genetic algorithm (GA), Grey Wolf Optimization (GWO), Beetle Antennae Search (BAS), Bayesian Optimization (BO), Particle Swarm Optimization (PSO) and Optina to fine tune its hyperparameters. The objective of the present work is to enhance the performance of ML model while ensuring robust generalization and computational efficiency. The performance of the tuned XGBoost models was assessed using evaluation metrics such as the coefficient of determination (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="42107_2025_1379_Article_IEq1.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$\hbox {R}^{2}$</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mtext>R</mtext> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation>), root mean square error (RMSE), and the time taken for optimization (in seconds). Among these hybrid models XGBoost-Optuna emerged as the best performing model while achieving the highest accuracy with testing <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="42107_2025_1379_Article_IEq1.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$\hbox {R}^{2}$</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mtext>R</mtext> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> (0.9345), minimal overfitting and the fastest optimization time (61.32 s). However, XGBoost-GWO demonstrated comparable accuracy for testing <InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="42107_2025_1379_Article_IEq1.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$\hbox {R}^{2}$</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mtext>R</mtext> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> of 0.9344 and generalization capability but required significantly higher optimization time (2821.25 s). XGBoost-Bayesian performed best against overfitting of the model but had a lower <InlineEquation ID="IEq4"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="42107_2025_1379_Article_IEq1.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$\hbox {R}^{2}$</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mtext>R</mtext> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> value of 0.9260 compared to other models. XGBoost-BAS offered a balanced trade off between accuracy and optimization time but did not outperform machine learning model optimized by Optuna.Overall, XGBoost-Optuna proved to be the most optimal choice by offering an excellent balance of accuracy, generalization, and computational efficiency which makes it a robust solution for predictive modeling in concrete mix design. </p>

Optimization-driven XGBoost model with metaheuristic algorithms for assessing compressive strength of high-performance concrete

期刊论文
发表日期：

2025年6月3日

查看全文

Amit Kumar Rai,
Shiv Shankar Kumar

摘要

Compressive strength (CS) is a key property of concrete mix, but the determination of CS requires costly and time intensive experimental procedures. Leveraging machine learning (ML) techniques for CS prediction can enhance accuracy and reliability while reducing the extensive need for laboratory testing.This study utilizes high performance concrete mix data set and employs the Extreme Gradient Boosting (XGBoost) ML model coupled with six metaheuristic optimization techniques such as genetic algorithm (GA), Grey Wolf Optimization (GWO), Beetle Antennae Search (BAS), Bayesian Optimization (BO), Particle Swarm Optimization (PSO) and Optina to fine tune its hyperparameters. The objective of the present work is to enhance the performance of ML model while ensuring robust generalization and computational efficiency. The performance of the tuned XGBoost models was assessed using evaluation metrics such as the coefficient of determination ( $\hbox {R}^{2}$ $R^{2}$ ), root mean square error (RMSE), and the time taken for optimization (in seconds). Among these hybrid models XGBoost-Optuna emerged as the best performing model while achieving the highest accuracy with testing $\hbox {R}^{2}$ $R^{2}$ (0.9345), minimal overfitting and the fastest optimization time (61.32 s). However, XGBoost-GWO demonstrated comparable accuracy for testing $\hbox {R}^{2}$ $R^{2}$ of 0.9344 and generalization capability but required significantly higher optimization time (2821.25 s). XGBoost-Bayesian performed best against overfitting of the model but had a lower $\hbox {R}^{2}$ $R^{2}$ value of 0.9260 compared to other models. XGBoost-BAS offered a balanced trade off between accuracy and optimization time but did not outperform machine learning model optimized by Optuna.Overall, XGBoost-Optuna proved to be the most optimal choice by offering an excellent balance of accuracy, generalization, and computational efficiency which makes it a robust solution for predictive modeling in concrete mix design.