Machine Learning-Based Prediction of Polymer Properties Using Structure-Property Relationship Modeling.

Area of Science:

• Polymer Science
• Materials Science
• Computational Chemistry

Background:

• Machine learning (ML) accelerates polymer property prediction but faces challenges with nonlinear structure-property relationships (SPRs), data standardization, and computational efficiency.
• Current ML methods often lack interpretability, hindering a deep understanding of the molecular features governing polymer properties.

Purpose of the Study:

• To develop a comprehensive ML framework for accurate prediction of polymer properties and identification of SPRs.
• To address limitations in prediction accuracy, interpretability, and computational efficiency in current polymer design methodologies.
• To provide insights into the influence of molecular features on polymer properties through advanced explanation techniques.

Main Methods:

• Integration of data preprocessing, molecular descriptor and topological index-based feature extraction, and iterative feature selection.
• XGBoost predictive modeling with hyperparameter optimization using the Starfish Optimization Algorithm (SOA).
• Application of SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) for model interpretability.

Main Results:

• The proposed framework achieved high predictive performance on the PolyOne dataset, with R² values > 0.92.
• Excellent accuracy was demonstrated with Mean Absolute Error (MAE) < 0.08 and Root Mean Square Error (RMSE) < 0.12 for key polymer properties.
• SHAP and LIME analyses provided both global and local insights into the molecular drivers of polymer properties.

Conclusions:

• The developed ML framework offers a robust and efficient solution for predicting polymer properties and understanding SPRs.
• The study successfully balances predictive accuracy, computational efficiency, and model interpretability, facilitating accelerated polymer design.
• This work provides a practical tool for researchers and engineers in polymer science, enhancing the design process and fundamental understanding.