Enhancing Concrete Compressive Strength Prediction through Machine Learning Techniques

Recent advances in artificial intelligence and machine learning are reshaping the field of materials science by enabling predictive frameworks that simplify the design process, substantially reducing reliance on traditional, costly, and labor-intensive testing methods. This study explores the potential of four machine learning techniques: Deep Neural Networks, Random Forest, Light Gradient Boosting Machine, and eXtreme Gradient Boosting, in predicting the compressive strength of concrete. While many studies have applied machine learning to this problem using the same established dataset, a notable gap in the literature is the reliance on validation methods prone to optimistic bias. The primary contribution of this work is the application of a more methodologically rigorous evaluation framework to address this gap. We employ a nested cross-validation (CV) scheme, where the outer loop provides an unbiased measure of model generalization while the inner loop systematically fine-tunes hyperparameters. By mitigating the risk of overfitting, this robust approach provides a comprehensive and reliable benchmark of each algorithm's true performance. These insights offer practitioners trustworthy guidance for forecasting concrete compressive strength, ultimately contributing to more efficient and reliable construction processes.