Prediction of refractive index of ceramic material by machine learning

Abstract

Terahertz (THz) spectroscopy, which allows to establish an accurate relationship between the intensity of the passing THz beam and the porosity of a non-metallic sample is one of the promising and non-destructive methods for studying the porous structure of composite materials. In order to optimize and reduce the costs of conducting experiments on the creation and study of the porosity of a ceramic material based on hydroxyapatite (HA), machine learning and regression analysis methods are studied to obtain predictive models of optical properties. Various methods such as linear and polynomial regression, as well as decision tree regression and random forest methods were used in developing regression models. For each method, the following metrics were calculated: mean square error, determination coefficient, and mean absolute error in percent, which made it possible to evaluate the effectiveness of each method. The scikit-learn library was used to calculate the root-mean-square error as a criterion for assessing the effectiveness of the methods. The most promising machine learning method for further use was identified - regression based on decision trees, which showed the lowest error in predicted values and received the highest model score for the determination indicator. The obtained results allow us to conclude that machine learning methods have an advantage in the task of predicting the refractive index of ceramic materials.
The work was performed according to the Government research assignment for ISPMS SB RAS, subject number FWRW-2022-0002.

Speaker

Anastasia E. Rezvanova
Institute of Strength Physics and Materials Science SB RAS
Russia

Discussion

Ask question