Abstract
Ionic liquids (ILs) are salts with a wide liquid temperature range and low melting points and can be fine-tuned to have specific physicochemical properties by the selection of their anion and cation. However, having a physical synthesis of multiple ILs for testing purposes can be expensive. For this reason, an in-silico estimation of physicochemical properties is desired. The selection of these components is limited by the low precision offered by state-of-the-art predictive models. In this paper, we explore the prediction of melting points with clustering algorithms and a novel Neuroevolution approach. We focused our design on simplicity. We concluded that performing clustering analysis in a previous phase of the model generation improves the estimation accuracy of the melting point, which is validated in experimentation made in-silico
Publisher
Editorial Académica Dragón Azteca