<p>High-performance concrete is vital in modern construction for its strength and durability, but predicting its compressive strength (CS) is challenging owing to mix complexity and material variability. This exploration presents a framework that integrates ML and optimization for accurate strength predictions. A database of 460 distinct high-performance concrete mixes with the main features of water-to-binder ratio, aggregate ratio, and curing age was deployed for the training and testing of predictive schemes. The work is novel as its utilization and fine-tuning of two hybrid optimization algorithms, the Hippopotamus Optimization Algorithm (HOA) and the Seagull Optimization Algorithm (SOA). These were used to boost the productivity of two base models; the Multi-Layer Perceptron (MLP) and the Gradient Boosting Regression (GBR) models by developing hybrid models, MLHO, MLSO, GBHO, and GBSO. Based on the experimental outcomes, among the MLP-based hybrid models, the MLSO (MLP optimized by SOA) demonstrated the highest predictive accuracy, achieving an R² of 0.922 and an RMSE of 7.049 MPa during the testing phase. In contrast, within the GBR-based hybrid models, the GBSO (GBR optimized by SOA) model outperformed all counterparts, attaining an R² of 0.979 and an RMSE of 3.595 MPa. These results establish GBSO as the top-performing scheme for predicting the CS, exhibiting superior generalization capability and minimal prediction error. The recommended method offers a scalable, data-driven solution to high-performance concrete design challenges, providing insightful information on material optimization with reduced costs and enhanced sustainability.</p> Graphical abstract <p></p>