Machine learning assisted prediction explainability and genetic algorithm optimization of tensile strength in fiber reinforced polymer composites as a computational framework toward hybrid composite design

The design of fiber-reinforced polymer (FRP) composites involves complex interactions among reinforcement type, matrix selection, and manufacturing parameters, making exhaustive experimental optimization costly and time-consuming. This study develops an integrated machine-learning framework that combines predictive modelling, explainable artificial intelligence, and surrogate-based optimization for data-driven tensile-strength assessment and configuration design of FRP composites. This computational framework is motivated by, and intended for eventual application to, kenaf/basalt/flax/carbon hybrid laminates (Sects. 1–2); the results reported in this abstract, however, were obtained on a separate, representative case-study dataset described below, not on the targeted hybrid system. Four regression models artificial neural network (ANN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR)---were systematically benchmarked using a 10,000-record representative FRP composite dataset comprising glass, carbon, aramid, and basalt reinforcements and four thermoset resin systems. The models were evaluated using held-out testing and cross-validation, while SHapley Additive exPlanations (SHAP) were employed to quantify the contribution of material and processing variables. A genetic algorithm (GA) was subsequently coupled with the best-performing surrogate model to identify high-strength parameter combinations within predefined manufacturing bounds. The ANN achieved the highest predictive performance, with a test-set coefficient of determination of (R^{2}=0.996), root-mean-square error of 38.2 MPa, and mean absolute error of 29.8 MPa, marginally outperforming XGBoost ((R^{2}=0.995)). These values were obtained under a random 80/20 split of a dataset whose record-level origin could not be independently verified and which was screened only for exact duplicates; they describe performance on this case-study dataset and are not presented as independently verified generalization. SHAP analysis identified fiber type and resin type as the dominant contributors to tensile-strength predictions, followed by layer count and void content. The GA converged toward a carbon/epoxy configuration with 24 layers, 180 °C curing temperature, 0.75 fiber volume fraction, and 0.5% void content, yielding a surrogate-predicted tensile strength of approximately 4680 MPa; this configuration lies at or near the boundary of the search space on nearly every parameter, so it should be read as an indication of the direction the surrogate model favors within the tested bounds, not as an interpolated, well-supported optimum (Sect. 4.4). These findings demonstrate the potential of an explainable ML–optimization workflow to reduce the search burden associated with composite-material design. Importantly, the reported numerical results characterize the representative case-study dataset and should not be interpreted as experimentally validated performance of the proposed kenaf/basalt/flax/carbon hybrid laminate. Experimental fabrication and ASTM-standard mechanical characterization are required to validate and subsequently adapt the framework to the targeted hybrid composite system.