This record contains the complete reproducibility package supporting the manuscript "Formulation-holdout validation for reliable machine learning in replicated materials experiments: an AA6061 composite case study." The study evaluates 12 independently prepared AA6061 hybrid-composite specimens nested within four reinforcement loadings (0, 5, 10, and 15 wt.%) and contrasts replicate-holdout validation with formulation-holdout validation (FHV). The archive includes the specimen-level dataset, executable Python analysis, pooled out-of-fold predictions, full model-performance audit, Matern-5/2 GPR predictive-interval audit, held-out formulation means, full-data GPR grid, conservative lower-confidence-bound score, uncertainty-ranked candidate formulations, selected next-experiment loadings, programmatically reproduced analysis figures, software requirements, verification checks, and provenance documentation. No synthetic observations, interpolated labels, or data augmentation are used. The archive is intended to reproduce the reported validation and uncertainty analyses and to document the distinction between repeatability-level prediction at known formulations and design-level generalization to unseen formulations.
Obinna Onyebuchi Barah, Milon Selvam Dennison, Akpan Samuel Cletus· Zenodo (CERN European Organi...· 0 citations
This record contains the complete reproducibility package supporting the manuscript "Formulation-holdout validation for reliable machine learning in replicated materials experiments: an AA6061 composite case study." The study evaluates 12 independently prepared AA6061 hybrid-composite specimens nested within four reinforcement loadings (0, 5, 10, and 15 wt.%) and contrasts replicate-holdout validation with formulation-holdout validation (FHV). The archive includes the specimen-level dataset, executable Python analysis, pooled out-of-fold predictions, full model-performance audit, Matern-5/2 GPR predictive-interval audit, held-out formulation means, full-data GPR grid, conservative lower-confidence-bound score, uncertainty-ranked candidate formulations, selected next-experiment loadings, programmatically reproduced analysis figures, software requirements, verification checks, and provenance documentation. No synthetic observations, interpolated labels, or data augmentation are used. The archive is intended to reproduce the reported validation and uncertainty analyses and to document the distinction between repeatability-level prediction at known formulations and design-level generalization to unseen formulations.
Obinna Onyebuchi Barah, Milon Selvam Dennison, Akpan Samuel Cletus· Zenodo (CERN European Organi...· 0 citations
Executable Online Resource and reproducibility archive accompanying the manuscript “Interpretable Benchmarking of Machine Learning Models for Small Experimental Energy Systems: Balancing Accuracy, Complexity and Physical Meaning in Biomass Gasification Prediction.” The archive contains the six direct QRO-401 analyzer records retained as empirical provenance anchors and the 98-row workbook-derived performance scenario dataset used for methodological benchmarking. The 98 scenarios are derived analytical records and must not be interpreted as 98 independent physical gasifier experiments. The repository provides the fixed analysis seed (20260901), frozen model specifications, complete repeated cross-validation generator, automated data-lineage and target-proximity detector, leave-configuration-out transport tests, full fold- and repeat-level computational results, manuscript-result verification outputs, machine-readable Online Resource tables, supplementary information, manuscript-aligned figures, pinned software environment files, repository manifest, and SHA-256 checksums. Tier A and Tier B constitute the legitimate interpolation benchmarking feature sets. Tier C includes formula-proximal energy production and is retained strictly as a leakage and formula-recovery diagnostic rather than as a deployable prediction benchmark. The archive is intended to support independent inspection, computational reproduction, provenance auditing, and verification of the results reported in the associated manuscript.
Obinna Onyebuchi Barah, Abdulrazak Jinadu Otaru, Ige Bori et al.· Zenodo (CERN European Organi...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
