Reproducibility Package and Reconstructed Dataset for Machine-Learning Modeling of Sorption and Thermodynamic Properties in Barhi Dates

This repository provides the reconstructed dataset, machine-learning benchmark results, out-of-fold predictions, model configurations, fold assignments, classical-model comparisons, variable documentation, and executable reproducibility resources associated with a study on machine-learning modeling of equilibrium moisture content and thermodynamic sorption properties in Barhi dates (Phoenix dactylifera L.). The reconstructed analytical dataset comprises 360 records corresponding to 120 unique experimental conditions and three controlled computational replicates per condition. Variables include water activity, temperature, drying method, sorption process, equilibrium moisture content (Xe), net isosteric heat of sorption (Qst), differential entropy (ΔS), and Gibbs free energy (ΔG). The primary predictive evaluation uses condition-grouped cross-validation so that all reconstructed replicates belonging to the same experimental condition remain within the same fold. The repository includes the data and supporting files required to inspect the validation structure, reproduce the reported benchmark results, and independently examine the comparison between machine-learning and classical modeling approaches. The reconstructed dataset derives from published experimental information and should not be interpreted as 360 independent laboratory measurements. The three records associated with each experimental condition are controlled computational replicates. The repository documents this structure explicitly to support transparent reuse and interpretation. The package is journal-independent and is intended to support reproducibility, independent verification, methodological reuse, and future research.