Predicting the Experimental Emission Spectra of Fluorescent Organic Semiconductors by Ensemble Machine Learning Analysis.

Abstract

The development of efficient and sustainable organic semiconductors is crucial for modern power source technologies, as they have the potential to revolutionize the way we harness and utilize energy. For current study, the emission maxima (λ_E) of 450 organic semiconductors are collected to analyze by machine learning (ML) related Random Forest and gradient boosting regressors. It identifies HallKier, FPdensityMorgan, and SMR_VSA as key descriptors influencing model performance, enabling accurate prediction of λ_E in organic semiconductors. The results showed that these models were able to predict the λ_E of the organic semiconductors with high accuracy. Further analysis using SHapley Additive exPlanations (SHAP) values revealed that chemical similarity plays an important role to determine their experimental λ_E. Interestingly, the study found that the synthetic accessibility (SA) of the organic semiconductors, which refers to the ease with which they can be synthesized, ranged from 0 to 0.20. The highest SA was found to correspond to λ_E in the range of 350-370 nm, which is typically associated with ultraviolet (UV) to blue light emission. This finding suggests that organic semiconductors with high SA tend to have λ_E in the UV to blue region, which is important for applications such as OLEDs and OPVs.