Interpretable Machine Learning Framework to Predict the Glass Transition Temperature of Polymers.

Area of Science:

• Materials Science
• Computational Chemistry
• Polymer Science

Background:

• Glass transition temperature (Tg) is critical for polymer energy absorption applications.
• Traditional methods for determining Tg are slow and costly trial-and-error processes.
• Machine learning (ML) offers a data-driven approach to predict material properties.

Purpose of the Study:

• To develop an accurate ML model for predicting polymer glass transition temperature (Tg).
• To identify significant features influencing Tg using advanced ML techniques.
• To establish an adaptable framework for predicting other polymer properties efficiently.

Main Methods:

• Utilized a dataset of 7174 polymer samples.
• Employed Morgan fingerprint and molecular descriptors for polymer representation.
• Applied feature selection methods including variance thresholding, Pearson correlation, and recursive feature elimination.
• Trained and compared nine ML algorithms, optimizing hyperparameters.
• Evaluated models using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R-squared (R²).

Main Results:

• The Extra Tree Regressor model demonstrated superior performance in predicting Tg.
• Identified key features influencing the glass transition temperature through statistical ML and SHAP analysis.
• The developed framework showed high adaptability for predicting other material properties.

Conclusions:

• ML, particularly the Extra Tree Regressor, provides an efficient and accurate method for predicting polymer Tg.
• Feature importance analysis reveals critical factors governing Tg, aiding material design.
• The computational framework is scalable and cost-effective for broader materials property prediction.