Exploratory Machine-Learning Prediction of Hard-to-Remove Water within a Treated Bleached Hardwood Kraft Pulp Series

The study evaluates whether fiber-morphology descriptors can predict hard-to-remove water (HRW) within a single never-dried northern bleached hardwood kraft pulp treatment series. All data were obtained from experimentally measured values reported in peer-reviewed literature.

The five supervised HRW conditions and their morphology come from Barrios et al. (2025), Biotechnology for Biofuels and Bioproducts 18:48, doi.org/10.1186/s13068-025-02646-1. The package contains: data: the 69-observation descriptive dataset (used for PCA), the five-condition predictor matrix, value-level provenance for every supervised value, and a data dictionary. notebooks: six executed Jupyter notebooks (PCA, nested leave-one-out screening, WRV reference model, Ridge sensitivity, exact permutation tests, and predictor-uncertainty sensitivity analysis). outputs: all numerical results, including out-of-fold predictions, fold selections, Ridge hyperparameters, permutation results, and sensitivity analyses.