A computational framework for optimizing LECA content in concrete using ANN and RSM

• Developed a computational framework to optimize LECA content in concrete. • Compared ANN and RSM models for predicting compressive strength loss and density reduction. • ANN outperformed RSM with R² up to 0.993 for density reduction and 0.983 for compressive strength loss. • Performed sensitivity analysis to identify key input parameters. • For a specific mix design, ANN determined the optimal LECA content as 10.37 %. Incorporating lightweight expanded clay aggregate (LECA) reduces the density of concrete but often results in undesirable compressive strength loss, creating a key design trade-off. This study develops a data-driven framework to determine the optimal LECA content by balancing compressive strength loss (CSL) and density loss (DL). A dataset of 129 records compiled from 26 published studies was analyzed. Artificial neural networks (ANN) and response surface methodology (RSM) were independently developed and compared to evaluate predictive reliability. The ANN model demonstrated superior accuracy, achieving coefficients of determination (R²) values of 0.983 for CSL and 0.993 for DL. Using the ANN predictions, a multi-objective optimization approach was formulated to identify the LECA content that provides the best compromise between mechanical performance and weight reduction. The results offer practical guidance for the design of lightweight concretes in construction applications.