Machine Learning Models for Melting Point Prediction of Ionic Liquids: CatBoost Approach

Abstract

Using ionic liquids as phase changing materials is of particular interest in the context of heat storage. As a consequence, predicting accurately the melting point of ionic liquids is of capital importance as it is one of the most important thermophysical properties in this context. In this work we consider a data set composed of 2249 different ionic liquids, with a majority of imidazole or ammonium cation-based molecules. We present a free and easy-to-use melting point predictive algorithm built on the CatBoost algorithm, making strong use of molecular descriptors. Based on LASSO, we select the most relevant descriptors for the task at hand and compare the model with previous ones.