Concrete is a fundamental material in modern construction, widely valued for its durability, availability, and versatility. However, designing an optimal concrete mixture remains challenging due to the complex interactions among its components. Traditional methods often involve numerous experiments, which are time-consuming, resource-intensive, and heavily reliant on trial and error, making it difficult to accurately capture these relationships. To address these challenges, this study proposes a combined approach using Design of Experiments (DoE) and machine learning (ML) to optimize recycled aggregate concrete mixtures for maximum compressive strength ( \(f_{c}\) ). A Light Gradient Boosting Machine model was implemented using a dataset of 594 samples collected from the literature, with hyperparameters optimized for performance. The model achieved \(R^2\) scores of 98% and 85% for the training and testing phases, respectively, and RMSE values of 1.74 MPa and 4.59 MPa. Sensitivity analysis indicated that the model’s behavior aligns with known input–output relationships in concrete mix design. The optimized model was used to predict \(f_{c}\) for configurations generated through DoE, facilitating the identification of the optimal mix. Rather than conducting multiple experiments, this approach enables the selection of a single experiment corresponding to the predicted highest \(f_{c}\) . This methodology was validated through two experimental case studies. In Case Study 1, the predicted \(f_{c}\) was 54.49 MPa, while the experimental result was 41.67 MPa. In Case Study 2, the predicted and measured strengths were 50.42 MPa and 49.41 MPa, respectively. These results demonstrate that the proposed approach can reliably identify high- \(f_{c}\) mixes, reducing the number of experiments required by traditional DoE methods. Specifically, the combined DoE-ML approach necessitates only a single experiment corresponding to the predicted optimal mix. This study validates the effectiveness of the integrated DoE-ML framework in streamlining the concrete mix design process, significantly reducing experimental costs and time when sufficient data are available.
