Presentation Information

[7p-N403-9]Prediction of Solvent Properties by Machine Learning with COSMO-RS Descriptors

〇Itsuki Kawanami1, Hiroto Yokoyama1, Akiko Kumada1, Masahiro Sato1 (1.Univ. of Tokyo Engineering)

Keywords:

semiconductor,materials,material property

In this study, we incorporated thermodynamic property values obtained from COSMOtherm, based on COSMO-RS theory, as descriptors in machine learning models to improve the prediction accuracy of solvent properties such as surface tension, solubility, and boiling point. Compared to conventional methods using quantum chemical descriptors, our approach demonstrated higher accuracy and superior extrapolation performance for unknown solvents, highlighting its potential as a promising tool for novel solvent discovery.