Machine learning prediction of mechanical properties in date palm fiber-reinforced PBS biocomposites under UV aging

The mechanical performance of date-palm-fiber-reinforced polybutylene succinate (DPF/PBS) biocomposites depends on several interacting material and aging parameters, making their prediction through conventional approaches challenging. This study develops a data-driven framework for predicting Young’s modulus (E), tensile strength (σ), and strain at break (ε) using fiber type, fiber content, treatment condition, and UV-aging duration as input descriptors. Four regression approaches were systematically evaluated: Linear Regression (LR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR). Model robustness was assessed using five-fold cross-validation, with predictive performance reported as mean ± standard deviation across the five folds and corresponding 95% confidence intervals. Feature-importance, feature-ablation, error, and computational-efficiency analyses were further performed to assess model interpretability and reliability. The results indicate that XGBoost provided the best overall predictive performance, achieving cross-validated R ² values of 0.971 ± 0.008, 0.814 ± 0.052, and 0.966 ± 0.030 for E, σ, and ε, respectively, with a marginal advantage over RF while requiring substantially lower training and prediction costs. SVR showed lower predictive performance, particularly for ε, for which its RMSE reached 11.89 ± 8.24%, compared with 4.67 ± 2.34% for XGBoost. Feature-importance and ablation analyses identified fiber content as the dominant predictor, while ultraviolet (UV)-aging duration showed a particularly important contribution to strain at break. The combined machine-learning and experimental approach demonstrates the potential of interpretable machine learning to predict the mechanical response of DPF/PBS biocomposites and provides a practically relevant tool for durability-aware material screening and optimization.