Predicting refractive index of inorganic compounds using machine learning

Area of Science:

• Materials Science
• Computational Chemistry
• Optics

Background:

• Refractive index (RI) is a critical optical property for materials.
• Accurate RI prediction is essential for material design and application.
• Existing methods for RI estimation can be computationally intensive.

Purpose of the Study:

• To develop a cost-effective machine learning model for predicting material refractive index.
• To identify key predictors for accurate RI estimation.
• To compare the performance of various regression algorithms for RI prediction.

Main Methods:

• Utilized experimentally measured RI values for 272 inorganic compounds.
• Employed band gap and atomic properties as predictor features.
• Investigated feature sets with 1, 5, 10, and 21 predictors.
• Compared six regression methods: OLSR, GPR, SVR, RFR, GBTR, and ERTR.

Main Results:

• Extremely randomized trees regression (ERTR) demonstrated the highest prediction accuracy.
• The machine learning model provides accurate RI estimations across a wide range.
• The model's prediction strength surpasses traditional empirical relations.

Conclusions:

• Machine learning offers a powerful and efficient approach for evaluating material refractive index.
• Band gap and atomic properties are significant predictors for RI.
• ERTR is a highly effective algorithm for RI prediction using this approach.