Exploratory Machine-Learning Prediction of Hard-to-Remove Water within a Treated Bleached Hardwood Kraft Pulp Series
Description
Reproducibility package for the manuscript "Exploratory Machine-Learning Prediction of Hard-to-Remove Water within a Treated Bleached Hardwood Kraft Pulp Series" (submitted to RSC Advances). 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, https://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. Length-weighted fiber length was selected in all five outer folds (R² = 0.807; exact permutation p = 0.025), but the result is not robust to the reporting precision of the source data and should be interpreted only as exploratory. To reproduce the results, install the packages in requirements.txt and run the notebooks locally or in Google Colab. See README.txt for details.
Files
Steps to reproduce
Download all files, keeping the folder structure (data, notebooks, outputs). Install Python 3.12 or later and the packages listed in requirements.txt (pip install -r requirements.txt jupyter). Run the six notebooks in the notebooks folder in order (01 to 06), either locally or in Google Colab. When run locally, each notebook finds the data automatically; in Colab, upload COMPLETE_DATASET.csv when prompted (notebook 06 also needs HRW5_VALUE_LEVEL_PROVENANCE.csv). Each notebook writes its results to a new folder inside the notebooks folder. Compare them with the archived results in the outputs folder. All analyses are deterministic or use fixed random seeds, so results reproduce exactly. See README.txt for a description of every file.
Institutions
- University of DhakaDhaka Division, Dhaka