Cross validated small data machine learning for composition property analysis of jute/prosopis juliflora bark/kenaf hybrid polyester composites
Machine-learning prediction from small experimental datasets is challenging. This study reanalyses a previously reported five-formulation jute fibre (JF)/Prosopis juliflora bark (PJb)/kenaf fibre (KF) hybrid polyester composite dataset (JF fixed at 50 wt%, PJb and KF varied complementarily) using fourteen regression algorithms under leave-one-out cross-validation (LOOCV), feature-importance analysis, PCA, t-SNE, and k-means clustering. A data-alignment error was identified and corrected prior to modelling. All fourteen algorithms returned negative LOOCV R² for tensile strength (best: SVR-RBF, R² = -0.376), reflecting the limited formulation count, the single effective compositional degree of freedom, and the peaked tensile response; impact strength, by contrast, was well predicted by ridge regression (R² = 0.883, MAE = 1.34 J/m²). Mechanically, increasing KF content from specimen A to E raised flexural modulus by 37%, flexural strength by 17%, hardness by 17%, and impact strength by 50%, while tensile strength and modulus peaked at the balanced KF25/JF50/PJb25 composition (specimen C). These contrasting outcomes show that ML reliability for small composite datasets is a joint function of sample size, effective dimensionality, and response shape, not sample size alone. Expanding future designs to 15–20 independently varied formulations with replicate measurements is recommended as a practical next step for statistically defensible predictive modelling in this material class.
Authors
- S. Sathees Kumar (ORCID: https://orcid.org/0000-0002-1294-8363)
- P. Sundaravadivel
- R. Muthalagu
Institutions
- Saveetha University (IN)
Publication Details
- Journal
- Discover Materials
- Published
- 2026-11-16
- DOI
- https://doi.org/10.1007/s43939-026-01040-4
- Primary Topic
- Natural Fiber Reinforced Composites
- Type
- article
- Field-Weighted Citation Impact
- 0.00