Machine Learning-Driven Characterization of Optical Materials: Predicting JO Parameters in Rare-Earth Doped Glasses

• Published in: Chemical Review and Letters
• Main Author: Singh J.P.; Newaz A.A.H.; Shirke M.B.; Humanante P.M.T.; Lee M.D.; Pandey V.K.
• Format: Article
• Language: English
• Published: Iranian Chemical Science and Technologies Association 2025
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85215939814&doi=10.22034%2fcrl.2024.488643.1474&partnerID=40&md5=b85ab45849a8633605d3963ea72ce0cf

This paper presents a machine learning-driven approach for predicting the spectroscopic properties of rare-earth (RE) doped glass systems, with a focus on Dy3+ ions. Glass compositions of 0.25 PbO–0.2 SiO2–(0.55−x) B2O3–x Dy2O3 were synthesized using the melt-quenching technique, and their density, molar volume, and Judd–Ofelt (JO) parameters (Ω2, Ω4, Ω6) were experimentally determined. The Judd–Ofelt theory was applied to calculate spectroscopic parameters such as oscillator strengths, radiative transition probabilities, and radiative lifetimes for Dy3+ doped glasses. Furthermore, a Random Forest (RF) regression model was developed to predict these parameters based on the composition of the glass. The model showed high accuracy, with R² (Coefficient of Determination) values above 0.9 and root-mean-square errors (RMSE) under 0.1, validating the use of RF for reliable predictions of optical properties. The results indicate that the RF model can effectively simulate the luminescent properties of Rare earth (RE)-doped glasses, significantly reducing the need for experimental testing. This approach offers potential for optimizing the design of optical materials used in applications such as lasers, optical amplifiers, and temperature sensors. © 2025, Iranian Chemical Science and Technologies Association. All rights reserved.