Abstract The construction industry's intensive resource use and waste generation necessitate sustainable alternatives. Using recycled coarse aggregates in concrete offers a viable solution. Accurately predicting the splitting tensile strength (STS) of such concretes through data‐driven methods reduces experimental needs, conserves materials, and enhances design efficiency and cost‐effectiveness. With this aim, this research thoroughly investigates the feasibility of optimized machine learning (ML) algorithms through particle swarm optimization (PSO) for modeling the STS of eco‐friendly steel fiber‐reinforced recycled coarse aggregate concrete (SRCAC). Accordingly, based on experimental data in the literature, a comprehensive and reliable database consisting of 335 STS values with 11 input features was adopted. Additionally, SHapley Additive exPlanations (SHAP)‐based feature importance and sensitivity analyses were implemented to make the “black box” behavior of PSO‐ML models more transparent and to provide insight into the model's decision‐making process. Given the results obtained, the PSO‐ML models developed in this paper demonstrate high accuracy, stability, and generalizability in data‐driven forecasting of STS of SRCAC, proving to be an effective tool for engineering applications. Specifically, the PSO‐CatBoost and PSO‐gradient boosting machine (GBM) models stand out with their high R 2 values (0.931 and 0.894, respectively) and low root mean squared error (RMSE) values (0.58 and 0.73, respectively) on the testing set, demonstrating that these models can be used as reliable decision support systems, thanks to their prediction accuracy as well as their low variance and balanced distribution. In addition, based on SHAP‐based explainability analyses, features such as the water/binder ratio (W/B), fiber factor (F), fine aggregate content (FA), and fiber length ( l f ) were determined as the most decisive parameters on STS, whereas the relative effect of variables like fiber volume fraction ( V f ) and fiber aspect ratio ( l f / d f ) remained limited. All in all, the PSO‐ML framework introduced makes it possible to forecast the STS value of SRCAC with high accuracy without the need for experimental testing, providing an important digital tool for sustainable and economical concrete design. These models not only reduce the use of natural resources by determining the optimal design parameters for concrete mixes containing recycled materials but also support environmental sustainability by promoting the reuse of construction waste and reducing the carbon footprint. Thus, data‐driven decision‐making processes enable the development of more effective, transparent, and eco‐friendly design strategies in civil engineering applications.
Utu et al. (Tue,) studied this question.