Prediction of Concrete Compressive Strength Using Support Vector Machine Regression: Statistical Characterization, Model Performance, and Feature Importance Analysis

Abstract

This study presents the development and evaluation of a machine learning–based framework for predicting the compressive strength of concrete using a Support Vector Machine (SVM) regression model. A comprehensive dataset comprising 1,133 concrete mix designs was employed, incorporating key material composition parameters, including cement, blast-furnace slag, fly ash, water, super-plasticizer, fine aggregate, coarse aggregate, and curing age. Prior to model development, extensive descriptive and advanced statistical analyses were conducted to examine the distributional characteristics, variability, skewness, and robustness of the input variables, ensuring a sound understanding of the data structure. The analysis revealed substantial variability in cementitious materials and curing age, highlighting the nonlinear and heterogeneous nature of concrete strength development. An ε-SVM regression model with a radial basis function kernel was implemented to capture these complex relationships. Model performance was assessed using an 80:20 train–test split and multiple statistical metrics, including mean squared error, root mean squared error, mean absolute error, mean absolute percentage error, coefficient of determination, and coefficient of variation of RMSE. The results demonstrate that the SVM model achieved strong predictive accuracy, with an R² value of 0.89, RMSE of 5.32 MPa, and MAPE of 11.38%, indicating reliable generalization to unseen data. Error analysis confirmed stable prediction behavior for most samples, with only a limited number of isolated outliers. Feature importance evaluation using univariate regression and a Relief-based algorithm identified cement content and curing age as the most influential parameters, followed by super-plasticizer and water content, in agreement with established concrete technology principles. Overall, the study confirms the suitability of SVM regression for concrete compressive strength prediction while emphasizing the importance of thorough data characterization, multi-metric evaluation, and feature interpretability for robust and physically consistent machine learning applications in civil engineering.