Machine learning-based model for prediction of concrete strength

The compressive strength of concrete is a crucial factor in the design and safety evaluation of reinforced concrete structures. Accurate and timely prediction of this property can significantly reduce costs and testing time. This study aims to overcome the limitations of traditional, labour-intensive laboratory tests by using advanced machine-learning techniques to improve prediction accuracy. A dataset from open-access platforms was enhanced using various data transformation methods to ensure reliability. Six widely used machine learning algorithms were implemented: Random Forest (RF), Decision Tree (DT), XGBoost (XGB), CatBoost (CB), Extra Tree Regressor (EXT R), and Light Gradient Boosting Model (LGBM), along with a Stacked Meta-Model ensemble. The models were evaluated using four statistical metrics: Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and R-squared (R²). XGBoost showed the best predictive accuracy in terms of MAE (3.45 MPa), outperforming the other models individually. However, CatBoost achieved superior results in terms of MAPE (6.23%) and R² (0.92), indicating better model fit and error minimization. The Stacked Meta-Model further improved overall performance, achieving the lowest MAPE (5.95%) and the highest R² (0.94), proving that the ensemble approach outperformed the individual models. Decision Tree exhibited the poorest performance with an MAE of 5.78 MPa, MAPE of 12.64%, and R² of 0.79, indicating its limited suitability for this task. RF, LGBM, and EXT R showed competitive results, with MAEs of 3.89 MPa, 3.76 MPa, and 3.69 MPa, respectively, and R² values close to 0.90. The study concludes that the Stacked Meta-Model consistently outperformed all individual models in predicting concrete compressive strength. A user-friendly Graphical User Interface (GUI) was also developed to enable practitioners to input relevant parameters and obtain accurate strength predictions efficiently.