Accurate prediction of concrete compressive strength is essential in construction and material science. Traditional laboratory methods for measuring compressive strength rely on destructive testing, which can be expensive and slow. These limitations create hurdles in scenarios requiring frequent strength evaluations. This study explores the effectiveness of ensemble machine learning models for predicting concrete compressive strength, combining predictions from multiple base models: Linear Regression (LR), Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting Regressor (GBR), XGBoost, and Multi-layer Perceptron (MLP). Using a dataset of 587 samples that covers a wide range of mix compositions and sand types, the ensemble model achieved superior predictive accuracy over individual models. Specifically, the ensemble model attained an <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="41939_2025_812_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="21" /> </InlineMediaObject> <EquationSource Format="TEX">$R^2$</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> score of 0.9876, Mean Squared Error (MSE) of 2.334, Mean Absolute Error (MAE) of 1.137 MPa, and Mean Absolute Percentage Error (MAPE) of 0.050, surpassing the top-performing base models (GBR and XGBoost). A SHAP (SHapley Additive exPlanations) analysis offered insight into feature importance, indicating that AGE, SP (Superplasticizer), and Binder Kg/m3 are key factors. These findings align with established knowledge about how curing age and binder content affect concrete strength. To encourage practical use of the trained model, a user-friendly Graphical User Interface (GUI) was developed, allowing real-time predictions for different mix designs. Hosted on GitHub, this interface can assist engineers and researchers in optimizing concrete mixes. This study demonstrates that an ensemble learning approach can offer accurate, interpretable, and accessible solutions for compressive strength prediction, addressing key limitations of existing laboratory testing methods. Future work could expand on this foundation by integrating additional data sources and exploring advanced feature engineering techniques to further enhance prediction accuracy and generalizability.

Ensemble machine learning models for predicting concrete compressive strength incorporating various sand types

期刊论文
发表日期：

2025年3月14日

查看全文

Rupesh Kumar Tipu,
Shweta Bansal,
Vandna Batra,
Suman,
Gaurang A. Patel

摘要

Accurate prediction of concrete compressive strength is essential in construction and material science. Traditional laboratory methods for measuring compressive strength rely on destructive testing, which can be expensive and slow. These limitations create hurdles in scenarios requiring frequent strength evaluations. This study explores the effectiveness of ensemble machine learning models for predicting concrete compressive strength, combining predictions from multiple base models: Linear Regression (LR), Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting Regressor (GBR), XGBoost, and Multi-layer Perceptron (MLP). Using a dataset of 587 samples that covers a wide range of mix compositions and sand types, the ensemble model achieved superior predictive accuracy over individual models. Specifically, the ensemble model attained an $R^2$ $R^{2}$ score of 0.9876, Mean Squared Error (MSE) of 2.334, Mean Absolute Error (MAE) of 1.137 MPa, and Mean Absolute Percentage Error (MAPE) of 0.050, surpassing the top-performing base models (GBR and XGBoost). A SHAP (SHapley Additive exPlanations) analysis offered insight into feature importance, indicating that AGE, SP (Superplasticizer), and Binder Kg/m³ are key factors. These findings align with established knowledge about how curing age and binder content affect concrete strength. To encourage practical use of the trained model, a user-friendly Graphical User Interface (GUI) was developed, allowing real-time predictions for different mix designs. Hosted on GitHub, this interface can assist engineers and researchers in optimizing concrete mixes. This study demonstrates that an ensemble learning approach can offer accurate, interpretable, and accessible solutions for compressive strength prediction, addressing key limitations of existing laboratory testing methods. Future work could expand on this foundation by integrating additional data sources and exploring advanced feature engineering techniques to further enhance prediction accuracy and generalizability.