Data-driven modeling of water adsorption isotherms in cocoa beans: Dataset and Python-based Machine Learning tools...

Area of Science:

• Food Science and Technology
• Agricultural Engineering
• Computational Science

Background:

• Cocoa bean quality is highly sensitive to moisture content during post-harvest storage.
• Understanding hygroscopic behavior is crucial for preventing spoilage and extending shelf-life.
• Existing data and modeling tools for cocoa moisture dynamics are limited.

Purpose of the Study:

• To generate a comprehensive dataset of water adsorption isotherms for dried and roasted cocoa beans.
• To develop and provide Python-based machine learning tools for multivariate modeling of cocoa moisture.
• To establish a framework for predicting moisture behavior and assessing storage risks in cocoa.

Main Methods:

• Experimental determination of water adsorption isotherms using the Dynamic Dewpoint Isotherm (DDI) method.
• Data collection across a range of water activity (0.1-0.85) and temperatures (25, 30, 40°C).
• Development of Python scripts implementing machine learning models (SVM, RF, ANN) for prediction.

Main Results:

• A detailed dataset of equilibrium moisture content (Xe) for dried and roasted cocoa beans under various conditions.
• Validated Python ML tools capable of predicting Xe based on water activity, temperature, and processing type.
• Demonstrated a reproducible and traceable workflow for data analysis and modeling.

Conclusions:

• The integrated dataset and ML tools provide a robust framework for managing cocoa moisture during storage.
• Reliable estimation of moisture-related parameters supports informed decision-making for quality preservation.
• This resource aids researchers, producers, and stakeholders in optimizing storage and minimizing post-harvest losses.