Prediction of Concrete Compressive Strength Using Ensemble Machine Learning Models Optimized by Lyrebird Optimization Algorithm and Human Evolutionary Optimization Algorithm

Precise prediction of concrete compressive strength plays a crucial role in designing concrete mixtures and assessing their quality. In this study, optimized ensemble machine learning models are developed and evaluated based on the UCI Concrete Compressive Strength dataset that are collected from laboratory tests on concrete mixtures. The predictor variables include cement, blast furnace slag, fly ash, water, superplasticizer, coarse aggregate, fine aggregate, and curing age, while the response variable denotes the concrete compressive strength. Two ensemble regressors, namely the extra trees regressor (ETR) and extreme gradient boosting regressor (XGBR), are optimized with the Lyrebird optimization algorithm (LOA) and the human evolutionary optimization algorithm (HEOA). The particle swarm optimization (PSO) and genetic algorithm are considered as comparison metaheuristics. A repeated holdout validation scheme is adopted, dividing the dataset into training, validation, and test sets at a ratio of 70:15:15. The validation set is used for selecting hyperparameters guided by the optimizer, while the test set is exclusively used to evaluate the final performance of the model. The nested cross-validation approach is not applied to optimize our proposed models. In this study, optimized XGBR-based models show better predictive performance than the ETR-based models. In particular, XGBR + PSO performed the best, with an R 2 value of 0.9486 ± 0.0182 and an RMSE of 3.7756 ± 0.6602 MPa. Among the LOA- and HEOA-based models, XGBR+HEOA obtained the best performance, with an R 2 value of 0.9483 ± 0.0174 and an RMSE of 3.7928 ± 0.6860 MPa.