A Machine-Learning Framework for Biodegradation Prediction in Sustainable Polymer Systems

Biodegradable polymers are increasingly central to sustainability-driven materials design, yet predictive modelling of their degradation behaviour across chemically diverse systems remains limited. While recent machine-learning studies have examined specific polymer classes or environment-dependent degradation scenarios, unified and interpretable cross-family frameworks are still underdeveloped. Predicting comparative biodegradation trends is particularly challenging due to nonlinear structure–property interactions and heterogeneous experimental reporting. In this study, a curated dataset comprising 90 samples from widely used biodegradable polymers including PLA, PCL, PHB, PBS, and representative blends was assembled to enable cross-family analysis under realistic data-scarce conditions. A unified machine-learning framework integrating hybrid chemical, thermal, mechanical, and structural descriptors was developed and systematically benchmarked against baseline linear and tree-based models. Among the evaluated approaches, XGBoost demonstrated superior and stable predictive performance. Robustness was assessed using five-fold cross-validation, bootstrap-based uncertainty quantification, and leave-one-polymer-family-out (LOFO) validation, providing explicit evaluation of descriptor-driven transferability beyond random splits. To enhance interpretability, a composite Hydrolysis Index was introduced to integrate hydrolysable bond density with diffusion-related constraints in a physically motivated manner. Explainable-AI analysis identified this index as the dominant predictive descriptor, while crystallinity, density, and stiffness-related parameters contributed meaningful secondary effects. Sensitivity analysis confirmed the bounded robustness of the descriptor formulation. A hybrid three-dimensional response-surface framework further visualized nonlinear descriptor interactions, offering mechanistically informed guidance for sustainable polymer design. Collectively, this work establishes a reproducible, interpretable, and cross-family modelling framework for biodegradation trend prediction, demonstrating how materials informatics can support rational screening, comparative analysis, and circular-economy-oriented polymer development. The proposed approach aligns with United Nations Sustainable Development Goals SDG 9, SDG 12, and SDG 13, providing a transparent foundation for accelerating sustainable materials innovation