Compressive Strength (CS) and Flexural Strength (FS) cannot be overstated when designing reinforced concrete structures, as industry standards govern these factors. Creating reliable predictive models for these properties offers the potential to enhance cost-effectiveness and efficiency by decreasing the necessity for numerous rounds of lab testing and iterative experiments, which are essential for obtaining valuable design information. In construction, evaluating the characteristics of advanced high-performance concrete (HPC) is crucial. Machine learning techniques are widely recognized as competent tools for forecasting the mechanical characteristics of concrete substances. This research centers on the comparison of different algorithms, including the standalone Catboost Regression model (CAT), hybrid models that incorporate the CAT model along with two distinct optimization algorithms, namely Artificial Rabbits Optimization (ARO) and Honey Badger Algorithm (HBA), as well as an ensemble approach that combines CAT with the ARO-HBA ensemble optimization algorithm. Considering the results indicated by the R2 values, the CAHB model showed significantly better performance in predicting both CS (R2trainCAHB=0.996) and FS (R2trainCAHB=0.994). Furthermore, when examining statistical accuracy metrics such as RMSE and MAE, it became evident that CAHB performed significantly better than CAT in forecasting the mechanical properties of concrete.