Data-driven optimized XGBoost and LightGBM for predicting the Judd-Ofelt parameters in rare-earth doped phosphate-based glasses

Advanced machine learning models, and more specifically the XGBoost and LightGBM models with Grey Wolf Optimization (GWO), can be utilized towards the prediction of the Judd-Ofelt (JO) parameters in rare-earth doped phosphate-based glasses. The JO parameters (${\Omega }_2$, ${\Omega }_4$, and ${\Omega }_6$) are critical for evaluating optical properties of rare-earth-doped materials, which are essential in applications, such as solid-state lasers, optical amplifiers, and three-dimensional displays. Traditional methods for estimating JO parameters involve complex methodologies that are particularly challenging for nanomaterials and powders. The use of the GWO-optimized XGBoost and LightGBM models promotes a novel and efficient approach to predict these parameters with high accuracy. The results of this study demonstrate that the combined GWO-LightGBM model outperforms others, achieving the lowest Mean Absolute Error (MAE) of 0.640 and the highest R² of 0.972 for predicting ${\Omega }_2$, with comparable performance for ${\Omega }_4$ and ${\Omega }_6$. This innovative approach simplifies the parameter estimation methodology and significantly enhances predictive accuracy, offering a valuable tool for the development of advanced optical materials.