Machine Learning- and SHAP Assisted Experimental Investigation of Annona reticulata Bark Fibre–Prosopis juliflora Wood–Jamun Seed Powder Hybrid Epoxy Biocomposites

Abstract

An integrated experimental and computational framework was developed to investigate epoxy hybrid biocomposites reinforced with Annona reticulata bark fibre, Prosopis juliflora wood powder and Jamun seed powder. Six formulations, APJ1–APJ6 were fabricated by hand lay-up and evaluated for flexural, tensile, Shore-D hardness, impact and moisture-absorption behaviour. Thermal stability, viscoelastic response, functional groups, crystalline characteristics and fracture morphology were examined using TGA, DMA, FTIR, XRD and SEM, respectively. Among the investigated formulations, APJ5 exhibited the highest tensile and flexural strengths of 267.4 and 341.2 MPa, respectively, whereas APJ6 achieved the highest Shore-D hardness of 85.2, demonstrating property-dependent optimum composition. Machine-learning prediction was performed using ANN, RF, GBR and KRR under five-fold cross-validation. ANN exhibited the best overall predictive performance, achieving R² values of 0.99735, 0.99767, 0.99806 and 0.99573 for tensile strength, flexural strength, impact strength and Shore-D hardness, respectively. SHAP analysis was subsequently employed to interpret the selected ANN model. Jamun seed powder and Prosopis juliflora wood powder exhibited equal normalized predictive contributions of 50% each, reflecting their complementary compositional variation rather than independent causal effects. The combined experimental, ML and SHAP framework provides a systematic approach for evaluating composition-dependent performance and identifying promising hybrid epoxy biocomposites for lightweight engineering applications.