Wood Moisture Dataset and Algorithms for Wood-drying prediction

Description

Dataset_Wood.csv: The dataset contains 1,846 specimen–time records from 142 drying curves. Curves 1–71 correspond to Pinus sp., and curves 72–142 correspond to Eucalyptus sp. Each curve contains measurements at 13 drying times. Nine observations classified as critical recording errors have missing moisture values and were not interpolated, resulting in 1,837 valid moisture-content observations used in the analyses. 01 – Exploratory data analysis (run_eda_analysis) 02 – Page baseline with grouped cross-validation (run_final_page_gpr_parametric_uncertainty.zip) 03 – GPR learning of Page parameters (run_gpr_page_parameter_learning.zip) 04 – Final parametric uncertainty propagation (run_final_page_gpr_parametric_uncertainty.zip) * run_eda_analysis.zip Exploratory data analysis and dataset organization. Reads the experimental wood-drying dataset, converts moisture content to percentage units, assigns specimen species and curve identifiers, defines descriptive drying phases, and generates the organized dataset, summary statistics, and exploratory figures used in the study. * run_page_model_groupkfold.zip Page-model baseline and grouped cross-validation. Fits individual Page drying curves and evaluates fixed Page-model parameterizations using five-fold specimen-grouped cross-validation. The script generates fold assignments, Page-parameter estimates, out-of-fold moisture predictions, performance metrics, residual summaries, and diagnostic figures. * run_gpr_page_parameter_learning.zip Gaussian Process Regression learning of specimen-dependent Page parameters. Trains pooled and species-specific GPR models to predict the Page parameters from specimen descriptors and compares predefined covariance kernels, including RBF, Rational Quadratic, Matérn 3/2, and Matérn 5/2, with and without an additional White Noise component. Out-of-fold reconstructed moisture trajectories and model-comparison metrics are generated. *run_final_page_gpr_parametric_uncertainty.zip Final Page–GPR parametric uncertainty analysis. Implements the selected species-specific Rational Quadratic Page–GPR configuration and propagates uncertainty in the GPR-predicted Page parameters through the Page equation using a first-order approximation. The script evaluates time-dependent uncertainty intervals, empirical coverage, interval width, and uncertainty calibration using held-out specimens. The files include the curated experimental wood-drying dataset and the computational workflows for exploratory analysis, Page-model baseline evaluation, specimen-grouped cross-validation, Gaussian Process Regression of Page parameters, and propagation and evaluation of parameter uncertainty.

Steps to reproduce

Download the dataset and the four Python scripts and extract all .zip files into the same working directory. Keep the experimental dataset with the original filename: tanaka_dados_dissertacao_sem_criticos.csv. (Dataset_Wood.csv). Install Python 3 and the required packages: numpy, pandas, matplotlib, seaborn, scipy, scikit-learn, and joblib. For example: pip install numpy pandas matplotlib seaborn scipy scikit-learn joblib Run the scripts from the project root directory in the following order: python run_eda_analysis.py This script reads the experimental dataset, converts moisture content to percentage units, assigns species and curve identifiers, and creates the eda_outputs directory. The main file required by subsequent analyses is: eda_outputs/organized_experimental_curves.csv. Run: python run_page_model_groupkfold.py This script fits the Page-model baseline and performs five-fold specimen-grouped cross-validation using a fixed random state. It creates the page_model_outputs directory, including the fold assignments, individual Page-parameter estimates, and out-of-fold Page predictions required by subsequent analyses. Run: python run_gpr_page_parameter_learning.py This script evaluates pooled and species-specific Gaussian Process Regression models for learning specimen-dependent Page parameters. The predefined covariance functions include RBF, Rational Quadratic, Matérn 3/2, and Matérn 5/2, with and without an additional White Noise component. Results are written to gpr_page_outputs. Run: python run_final_page_gpr_parametric_uncertainty.py This script implements the selected species-specific Rational Quadratic Page–GPR configuration and propagates the GPR-predicted Page-parameter uncertainty through the Page equation using a first-order uncertainty-propagation approximation. Results are written to final_page_gpr_parametric_uncertainty_outputs, including prediction tables, uncertainty intervals, coverage metrics, figures, fitted models, and reproducibility reports. The scripts use relative file paths; therefore, they should be executed from the same project root directory and in the order indicated above. No interpolation of missing moisture observations is performed. The additional quality-control columns contained in the original CSV are retained in the shared dataset but are not required as predictors by the analysis scripts. As a reproducibility check, the selected species-specific Rational Quadratic Page–GPR model should yield an overall out-of-fold RMSE of approximately 4.407 percentage points, MAE of 2.197 percentage points, \(R^2\) of approximately 0.988, and mean prediction bias of approximately −0.235 percentage points. The propagated 95% uncertainty intervals for observations after day 0 should provide empirical coverage of approximately 95.75%.

Institutions

• Universidade Federal de Uberlândia

  Minas Gerais, Uberlândia

Categories

Funders

• Conselho Nacional de Desenvolvimento Científico e Tecnológico

  Ministry of Science, Technology and Innovation

  Federal District, Brasília

  Grant ID: 309248/2025-6

Licence