DefinePK

Abstract

This study investigates the predictive modeling of concrete compressive strength using advanced statistical and machine-learning techniques. A curated dataset of 1,030 concrete mixes including cement, blast-furnace slag, fly ash, water, superplasticizer, coarse and fine aggregates, and curing age was subjected to rigorous preprocessing and exploratory analysis. Pairwise correlations revealed negligible linear relationships with compressive strength, highlighting the inherently nonlinear interactions among mixture constituents. Five representative algorithms Ordinary Least Squares, Ridge, Lasso, Random Forest, and Gradient Boosting were trained with cross-validation and evaluated on an independent test set. All models exhibited low coefficients of determination and root-mean-square errors around 24 MPa, indicating that raw mixture proportions and age alone cannot adequately capture the complex hydration chemistry and microstructural evolution governing strength development. Feature-importance analysis identified supplementary cementitious materials and aggregate gradation as key contributors. The results emphasize the necessity of domain-informed feature engineering and provide a reproducible methodological framework for performance-based, data-driven concrete mix design