Robust Machine Learning for Predicting Thermal Stability of Metal-Organic Framework

This study introduces a quantitative structure-property relationship (QSPR) model to predict the thermal stability (TS) of zinc-based metal-organic framework (Zn-MOF) compounds using machine learning. The model achieves higher prediction accuracy by integrating robust regression (RR) to handle outliers effectively than traditional multiple linear regression (MLR). Validated through metrics like root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination (R²), the RR model demonstrates superior performance. Notably, the TS of a novel Zn-MOF, Zn₃(DDB)(DPE)·H₂O, was successfully predicted and experimentally validated, confirming the model’s reliability. This work addresses the challenge of accurately predicting TS in MOFs, underscoring the potential of machine learning in materials design.