AI model estimates how quickly a common bioplastic degrades
Researchers trained machine-learning models to forecast weight loss in PHBV, a biodegradable plastic, across soil, water, marine and compost settings.
By Tom Brennan · Health & Medicine Correspondent
3 min read
Researchers at the Agricultural University of Athens have built a machine-learning tool to estimate how fast a biodegradable plastic loses weight under different environmental conditions. The work matters because manufacturers and scientists need more than a biodegradability label; they need evidence on how materials behave in soil, water, marine settings and compost.
The study, published in Polymers, focused on PHBV, or poly(3-hydroxybutyrate-co-3-hydroxyvalerate), which the researchers describe as one of the most studied members of the polyhydroxyalkanoate, or PHA, family. PHAs are bio-based plastics made through microbial fermentation rather than oil refining, according to the ANIPH project.
The iCube Programme said the work was carried out by the Agricultural University of Athens, a partner in ANIPH, which is developing wound dressings and packaging based on materials intended to return safely to the environment after use.
Why degradation speed is hard to pin down
Plastic use has risen sharply over decades. According to figures cited by iCube, global plastic production grew from 1.5 million metric tons in 1950 to 359 million metric tons in 2018. The same account said more than 60% of household plastic waste consists of single-use food packaging made from petroleum-based plastics.
Biodegradable plastics are meant to reduce part of that burden, but their environmental performance depends on conditions. A material may break down at a different pace in compost than in seawater or freshwater, and its chemical makeup can also affect the process, according to the Athens team’s study.
To address that uncertainty, the researchers assembled a database of 1,467 time-based measurements tracking weight loss in PHBV samples. The measurements covered laboratory and real-environment tests, according to the paper by Marianna I. Kotzabasaki and colleagues.
Two machine-learning models tested
The team used Random Forest and XGBoost, two machine-learning methods, to train predictive models. The paper describes the approach as a QSAR model, which links a material’s characteristics and its surrounding conditions with measured weight loss during biodegradation.
The researchers then tested the models against data withheld from training. According to iCube, both models explained more than 92% of the variation in weight loss on test data and more than 96% on training data.
The analysis identified three leading predictors: exposure time, the degradation environment and the hydroxybutyrate ratio, a chemical-composition measure within the polymer. The environments assessed included soil, sea, fresh water and compost, according to the study summary.
The team also made the model available as a web application on the Jaqpot computational platform through the ANIPH virtual organization. According to iCube, the tool is intended to let researchers, manufacturers and other users estimate how a selected PHBV formulation is likely to degrade in a chosen environment.
The paper, “A Data-Driven Framework for Predicting PHBV Biodegradation-Induced Weight Loss Based on Laboratory and Real-Environment Condition Tests,” was published in Polymers in 2026. Its DOI is 10.3390/polym18070897.
This story draws on original reporting from Phys.org.