Metaheuristic-optimized gradient boosting models for predicting the compressive strength of WPP-modified high-performance concrete

This study presents a novel machine learning-based framework for predicting the compressive strength of high-performance concrete incorporating waste PVC powder (WPP) as a partial substitute for fine aggregate. A baseline Gradient Boosting Machine (GBM) model was developed and optimized using four advanced metaheuristic algorithms: the Whale Optimization Algorithm (WOA), the Dragonfly Optimization Algorithm (DOA), the Earthworm Optimization Algorithm (EOA) and the Adaptive Bat Algorithm (ABA). The experimental procedure is carried out with varying WPP content (0–15%) and curing ages (3–90 days) as per the Indian Standards. A dataset of a total of 189 instances is collected. Performance evaluation through 5-fold cross-validation demonstrated that the hybrid GBM-WOA model achieved superior predictive performance with a test R2 of 0.9949, RMSE of 0.0195 and MAPE of 1.82%. The standalone GBM model (R2 = 0.8004, RMSE = 0.1296, MAPE = 10.03%) performs lower than the hybrid models. Sensitivity analysis revealed that fine aggregate content, curing age and WPP had relative influences of 38%, 34% and 28%, respectively. The study confirms the effectiveness of metaheuristic-optimized GBM models in accurately modeling the complex nonlinear behavior of sustainable concrete materials and offers a fast and accurate data-driven tool to reduce reliance on extensive laboratory testing.