Machine learning-based prediction and SHAP interpretation of mechanical properties in Kenaf/PALF/flax reinforced vinyl ester hybrid composites

Hybrid Natural Fiber Composites (HNFCs) made of kenaf, PALF, and flax fiber reinforced vinyl ester are the sustainable alternative for engineering applications, but the mechanical properties of these composites are dependent on complex and nonlinear interactions between the fibers, which make traditional experimental optimization not feasible. This work aims to systematically compare and study four machine learning techniques viz., Linear Regression (LR), Ridge Regression (RR), K-Nearest Neighbors (KNN) and Support Vector Regression with Radial Basis Function kernel (SVR-RBF) to predict the tensile strength (67.51–73.67 MPa), flexural strength (157.06–187.31 MPa), impact strength (9–15 J/m²) and hardness (82–93 Shore-D) from the experimental data obtained from six hybrid formulations (HC1–HC6) reported in the literature. SVR-RBF model was the best for predicting the tensile strength (R² = 0.9975), flexural strength (R² = 0.9978), impact strength (R² = 0.9895) and hardness (R² = 0.9454), whereas linear models failed in predicting impact strength (R² = 0.03). The results of the SHapley Additive exPlanations (SHAP) analysis indicated that the mechanical properties were mostly associated with the interactions of the hybrid fibers, with PALF–flax dominating tensile (SHAP = 0.48–0.50) and flexural properties, kenaf–PALF the impact (SHAP = 0.38–0.40) and hardness (SHAP = 0.95–1.00) properties, and the individual fiber contributions being smaller (SHAP < 0.16). The trained SVR-RBF model was then applied to estimate the mechanical properties of five untrained formulations (HC7–HC11), highlighting the promise for preliminary material screening and composition optimization using the proposed machine learning framework. These formulations have yet to be experimentally produced and verified and the reported values are therefore estimates and may not be accurate. The predictions indicate that the mechanical properties can be lowered for fiber compositions that are not balanced due to the decrease in hybrid fiber synergy. The proposed framework is fast, interpretable and cost-effective for the initial screening and design of sustainable hybrid composites for automotive and civil engineering applications.