Machine Learning for Predicting Performance and Degradability of Bio-Based Food Packaging

Abstract

As global demand for sustainable food packaging intensifies, bio-based materials have emerged as promising alternatives to petroleum-derived plastics. However, their adoption at industrial scale is hindered by inconsistencies in mechanical performance, barrier properties, and biodegradability under varying environmental conditions. This study proposes a machine learning (ML)-based framework to predict the performance and environmental behavior of bio-based food packaging materials based on their composition, processing parameters, and environmental variables. Multiple ML algorithms, including Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Networks (ANN), were trained using experimental datasets derived from published studies and lab-generated data. Key target variables included tensile strength, water vapor transmission rate (WVTR), oxygen permeability, and biodegradation time. Results demonstrate that ensemble models such as RF provided high prediction accuracy and interpretability, enabling efficient material screening and optimization. The study highlights the potential of data-driven tools to accelerate the development of eco-friendly packaging solutions with tailored properties, reducing trial-and-error experimentation and promoting circular economy objectives.