Machine learning in constructing structure–property relationships of polymers

The properties of polymer materials are closely related to their structures. A deep understanding of quantitative relationships between the structures and properties of polymers is crucial for the design and preparation of high-performance polymer materials. However, these relationships are inherently complex and difficult to model with limited trial and error experimental data. In recent years, machine learning (ML) has become an effective multidimensional relationship modeling method, playing an important role in the construction of quantitative relationships between the structures and properties of polymer materials. This review first provides an overview of the ML workflow, with a focus on the feature engineering of polymers and commonly used ML algorithms in the application of ML processes. Afterward, the progress of ML in the quantitative relationship between the structures and properties of polymer materials was summarized and evaluated from the aspects of mechanical properties, thermal conductivity, glass transition temperature (Tg), compatibility, dielectric properties, and refractive index of polymers. Finally, the application prospects of ML in polymer material research were proposed.