Predicting glass transition temperatures for OLED organics with random forest algorithm

Abstract

Organic light-emitting diodes (OLEDs) have attracted much attention because of their excellent performance advantages in color quality, viewing angle, flexibility and manufacture. Their glass transition temperatures (T_gs) directly determine the thermal stability and define the potential applications. In this work, after generation of 499 Dragon molecular descriptors, a quantitative structure–property relationship (QSPR) model was generated for correlating T_gs of 2091 OLED molecules with their Dragon descriptors, by applying random forest algorithm. After excluding two outliers, the best random forest model with a determination coefficient (R²) of 0.956 and a root-mean-square (rms) error of 12.85 K for the training set of 1880 OLED molecules, achieved test set R² of 0.850 and rms = 18.07 K for 209 OLED molecules. The random forest model suggested was found to be accurate in comparison with previous QSPR models reported on T_gs of OLED molecules.