Machine learning applications to predict the split tensile and flexural strengths of rubber tire-concrete composite

We evaluate the prediction model of the split tensile strength ( f t ) and flexural strength ( f b ) of a concrete mix with rubber tires as a partial replacement aggregate. Analyzing and utilizing historical data, we predict f t and f b based on the compressive strength ( f c ) and rubber tire concentration. The Pearson correlation matrices show that rubber tire concentration exhibits moderate-to-strong correlations with concrete mechanical properties (r = −0.72 for minimum f t treatment and r = −0.63 for minimum f b treatment). In contrast, mechanical properties demonstrate strong interdependency (r = 0.96 between fc control and maximum f b treatment). Three machine learning (ML) algorithms, namely decision tree (DT), Gaussian process (GP), and eXtreme gradient boosting (XGBoost), were compared for predictive performance. DT models identified fc strength as the primary decision criterion with threshold values of 19.95 MPa for f t and 34.5 MPa for f b prediction. In addition, the f t and f b GP models' predictions based on fc effectively quantified uncertainty through confidence intervals ranging from 65% to 95%, while capturing non-linear relationships. Furthermore, XGBoost models demonstrated the highest performance prediction with an R 2 of 0.9048 for f t (MAPE = 10.78%) and an R 2 of 0.9503 for f b (MAPE = 29.25%). These results substantially outperformed DT and GP models and were selected for validation. However, k-fold cross-validation revealed significant overfitting, with cross-validation performance dropping to mean R 2 values of −0.0126 for f t and 0.3436 for f b models. Despite excellent training performance, our findings indicate a moderate generalization capability of the XGBoost models. These results also establish XGBoost as the best-performing approach for predicting rubber tire-concrete composites. However, they illustrate the need for expanded datasets to enhance model generalization.