Machine learning to predict how fast biodegradable plastics break down in nature

plastic sea
Credit: Unsplash/CC0 Public Domain

Testing how quickly a biodegradable plastic actually breaks down in the environment can take months, sometimes years, of lab work. A new study from the Agricultural University of Athens, offers a faster alternative: a machine-learning tool that predicts biodegradation outcomes for a widely used bioplastic almost instantly.

The research, published in Polymers, focuses on PHBV (poly(3-hydroxybutyrate-co-3-hydroxyvalerate)), a biopolymer produced naturally by bacteria and considered a promising, non-microplastic-forming replacement for conventional fossil-based plastics, particularly valuable in settings like humanitarian crises where waste management infrastructure is limited.

The team, led by Chrysanthos Maraveas, built a curated database from 13 peer-reviewed studies spanning nearly three decades, capturing how PHBV formulations—with different additives, compositions and environmental conditions—degrade over time, measured through CO2 evolution (mineralization). The resulting data set covered 93 experimental instances and more than 1,300 individual biodegradation measurements.

Two machine learning approaches, Random Forest and XGBoost, were trained on this data and tested against unseen experimental instances. Both achieved strong predictive accuracy, with R2 values around 0.95–0.97 even on fully held-out data, meaning the models reliably generalize beyond the examples they were trained on.

Analysis of the models revealed that biodegradation time was, unsurprisingly, the strongest predictor, reflecting the fundamentally kinetic nature of the process. But temperature, the ratio of the polymer's two building blocks (hydroxyvalerate and hydroxybutyrate), the degradation mechanism (particularly surface erosion), microbial community type and additive content all played meaningful secondary roles, confirming that biodegradation is governed by a complex interplay of material design and environmental conditions rather than time alone.

The Random Forest model has been made publicly available as a free, interactive web tool on the Jaqpot platform, allowing researchers and manufacturers to input formulation and environmental parameters and receive rapid biodegradation predictions, supporting a "safe-and-sustainable-by-design" approach to developing next-generation biodegradable materials.

More information

Marianna I. Kotzabasaki et al, Machine Learning Methods for Mineralization-Based Biodegradation Prediction in Polyhydroxyalkanoate-Based Biopolymers: Insights from Lab-Scale Experiments, Polymers (2026). DOI: 10.3390/polym18091076

Who's behind this story?

Sadie Harley

Sadie Harley

BSc Life Sciences & Ecology. Microbiology lab background with pharmaceutical news experience in oil, gas, and renewable industries. Full profile →

Andrew Zinin

Andrew Zinin

Master's in physics with research experience. Long-time science news enthusiast. Plays key role in Science X's editorial success. Full profile →

Citation: Machine learning to predict how fast biodegradable plastics break down in nature (2026, July 7) retrieved 14 July 2026 from https://phys.org/news/2026-07-machine-fast-biodegradable-plastics-nature.html

This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.