Predicting compressive and tensile strength of concrete with different sand types using machine learning

Abstract

The study evaluates the prediction accuracy of concrete compressive and tensile strength using machine learning and deep learning models with diverse sand materials, cement types, and filler combinations. A dataset of 587 concrete mix samples was compiled from literature. Seven prediction algorithms, including Linear Regression, Support Vector Regression, Decision Tree Regressor, Random Forest Regressor, Gradient Boosting Regressor, Extreme Gradient Boosting, and Multi-Layer Perceptron, were employed. Preprocessing involved handling missing data, outlier detection, encoding, and feature normalization for Principal Component Analysis. Performance was assessed using coefficient of determination (R²), root mean squared error, mean absolute error, and mean absolute percentage error with cross-validation. Extreme Gradient Boosting showed superior accuracy (R² = 0.954 for compressive and 0.952 for tensile strength). SHapley Additive Explanations identified curing age, water-to-binder ratio, and filler density as key features. A real-time GitHub interface allows strength predictions, aiding sustainable construction optimization.