A key goal of environmental policies and circular economy strategies in the construction sector is to convert demolition and industrial wastes into reusable materials. As an industrial by-product, Waste marble (WM), has the potential to replace cement and fine aggregate in concrete which helps with saving natural resources and reducing environmental harm. While many studies have so far investigated the effect of WM on compressive strength (CS), it is undeniable that conducting experimental activities requires time, money, and re-testing with changing materials and conditions. Hence, this study seeks to move from traditional experimental approaches towards artificial intelligence-driven approaches by developing three models—artificial neural network (ANN) and hybrid ANN with ant colony optimization (ACO) and biogeography-based optimization (BBO) to predict the CS of WM concrete. For this purpose, a comprehensive dataset including 1135 data records is employed from the literature. The models’ performance is assessed using statistical metrics and error histograms, and a K-fold cross-validation analysis is applied to avoid overfitting problems, emphasize the models’ reliable predictive capabilities, and generalize them. The statistical metrics indicated that the ANN-BBO model performed best with a correlation coefficient (R) of 0.9950 and root mean squared error (RMSE) of 1.2017 MPa. Besides, the error distribution results revealed that the ANN-BBO outperformed the ANN and ANN-ACO with a narrower range of errors so that 98% of the predicted data points in the training phase by the ANN-BBO model experienced errors in the range of [-10%, 10%], whereas for the ANN-ACO and ANN models, this percentage was 85% and 79%, respectively. Additionally, the study employed SHapley Additive exPlanations (SHAP) analysis to clarify the impact of input variables on prediction accuracy and found that the specimen’s age is the most influential variable. Eventually, to validate the ANN-BBO, a comparison was performed with the results of previous studies’ models.
