Improved Prediction of Maximum EQE in TADF-based OLEDs Through Ensemble Learning

In recent times, we have seen colossal growth in the field of Thermally Activated Delayed Fluorescence (TADF)-based OLEDs in terms of synthesis and applications in sensing and imaging. However, the device-level application is still limited to the unpredictability of external quantum efficiency (EQE). Although theoretical research involving internal quantum efficiency (IQE) and mechanistic pathways for reverse intersystem crossing (rISC) in TADF systems have been explored quite rigorously, investigation on EQE is lacking. With the emergence of data-driven analysis being the fourth paradigm of science (empirical, theoretical, and computational being the previous three), we have employed ML models on 30 features of 123 samples, availed from literature to predict the EQEmax. On the one hand, the employed models capture device selectivity but are prevalent in the emissive range of chromophores. We have shown that Gradient Boosting (GB), an ensemble learning model, has been able to predict EQEmax with r2 score of 0.71?à0.04/0.84 and a low RMSE of 4.22?à0.55/2.53 for the train/test set. Considering the current state-of-the-art (SOTA), this is the best model which can predict for TADF chromophores of any emissive range and delineate the effect of device architecture. We also have carried out feature importance analysis to make this so-called black-box model interpretable. This analysis has helped to figure out essential parameters responsible for better EQE efficiency. Even the learning curve is still ascending, proving that the model can improve its prediction if more training examples are provided in the future. All the computations can be done using easily accessible cloud computations.