Discovering optical performance enhancers in perovskite materials through machine learning-based feature analysis

Perovskites attract significant attention as a coating material in optical fiber sensing, but challenges remain due to the limited discovery of suitable materials and the high trial-and-error costs, resulting in only a few perovskites being used in optical sensing experiments. Addressing this issue, a novel systematic computational screening strategy for perovskites is established. This strategy is demonstrated to accelerate the discovery of perovskite coating materials that can enhance optical sensing sensitivity. These perovskites are defined in this study as optical fiber performance enhancers (POPEs). For the most accurate prediction results, 10 sampling methods combined with 10 classification algorithms are compared. Following 100 comparative experiments, the model using the SMOTE-ENN sampling methods and the label spreading (LS) algorithms shows 100% accuracy and precision in leaving-one-out cross-validation (LOOCV). However, this result should be supported with further experiments and numerical simulations. Finally, we feed 500 samples of photonic, piezoelectric, ferroelectric, magnetic, and other perovskite materials into the optimal model, resulting in 237 potential POPEs for the first time. Meanwhile, we predicted the probabilities of forming POPEs using 10 perovskites commonly used in the field of fluorescence sensing. The obtained values of probability of forming POPEs are all over 91%, which indirectly validates our screening strategy for perovskites is effective. These 237 POPEs show promising prospects for becoming the forefront materials in the next generation of fiber optic sensing technologies.