Uncertainty Quantification in Machine Learning for Polymer Properties: A Benchmark Study

Machine learning (ML) has emerged as a transformative tool in materials science, enabling accelerated discovery and design of novel molecules while reducing experimental costs. Uncertainty quantification (UQ) is crucial for enhancing the reliability of ML predictions, particularly in high-stakes applications, such as functional polymer discovery. In this study, we present a comprehensive evaluation of six UQ methods in ML—Ensemble, Bayesian Neural Network (BNN), Gaussian Process Regression (GPR), Monte Carlo Dropout (MCD), Mean-Variance Estimation (MVE) and Evidential Deep Learning (EDL)—for predicting key polymer properties, including glass transition temperature (Tg), band gap (Eg), melting temperature (Tm) and decomposition temperature (Td). The models are assessed using three independent metrics, including prediction accuracy (R2), Spearman’s rank correlation coefficient and calibration area, offering a robust framework for evaluating both mean predictions and uncertainty estimates. Our analysis spans datasets of four properties, out-of-distribution (OOD) experimental and molecular dynamics (MD)-derived data, high-Tg polymers and diverse polymer types, providing a holistic perspective on model performance. BNN emerged as the most versatile model, achieving strong accuracy and reliable UQ across most scenarios. Ensemble models demonstrated superior performance for high-Tg polymers with significantly higher computational cost. Other methods, including GPR, MCD and MVE, showed varying levels of accuracy and reliability, with limitations in correlating predicted uncertainties with actual errors. This study highlights the critical role of UQ in guiding experimental validation and optimizing ML-driven material design workflows. By providing insights into model performance under diverse scenarios, these findings establish a benchmark for integrating ML models with UQ techniques in accelerating the discovery of advanced polymers.