Dataset and machine learning-based computer-aided tools for modeling working sorption isotherms in dried parchment and green coffee beans

This work presents a comprehensive dataset on working sorption isotherms and mid-infrared spectra for parchment husk, parchment coffee, and green coffee beans (Coffea arabica L.). The working sorption isotherms were experimentally determined using the Dynamic Dewpoint Isotherm (DDI) method, covering typical storage conditions of dried coffee beans in warehouses. These conditions account for a water activity range (a_w) from 0.1 to 0.9 and temperatures of 25°C, 35°C, and 45°C. Furthermore, the mid-infrared spectra of parchment husk and green coffee were obtained using Attenuated Total Reflectance-Fourier Transform Infrared (ATR-FTIR) spectroscopy as a complementary tool to analyze the role of parchment covering in the water sorption behavior of parchment coffee beans. The dataset also provides computer-aided tools for the mathematical modeling of working sorption isotherms and infrared data of coffee. These tools were developed using MATLAB® R2023a (The MathWorks Inc., Natick, MA, USA) and offer users the ability to model sorption isotherms and analyze infrared spectral data using advanced machine learning techniques. Thereby, the MATLAB scripts implement an automated routine for the calibration and optimization of the Support Vector Machine (SVM) and Random Forest (RF) techniques, enabling the modeling of working sorption isotherms for each coffee type (considering only a_w and temperature) and in a multivariate approach (incorporating a_w, temperature, and coffee type) to predict the equilibrium moisture content (X_e). Additionally, the MATLAB script for Principal Component Analysis (PCA) enables users to perform advanced chemometric modeling of the coffee spectra. This script provides a latent-variable-based tool for analyzing spectral patterns associated with different coffee types, allowing for robust model-based differentiation of coffee samples using their infrared properties. These models are particularly valuable as digital representations of the coffee storage process and can be used to optimize storage conditions, understand hygroscopic behavior, and ensure moisture-based quality monitoring in parchment and green coffee beans. The experimental dataset, including working sorption isotherms and mid-infrared spectra, is organized into Excel sheets according to experimental conditions and replicates. The MATLAB scripts come with ready-to-use computational instructions for calibrating predictive models, ensuring precise fitting of isotherms and spectral properties. This dataset represents a valuable asset for researchers, coffee producers, and industry stakeholders, providing practical tools for storage optimization, shelf-life determination, and in-depth analysis of water sorption behavior across different coffee processing stages.