Predicting Properties of Imidazolium-Based Ionic Liquids via Atomistica Online: Machine Learning Models and Web Tools
Stevan Armaković, Stevan Armaković, Sanja J. Armaković, Sanja J. Armaković
Abstract
Machine learning models and web-based tools have been developed for predicting key properties of imidazolium-based ionic liquids. Two high-quality datasets containing experimental density and viscosity values at 298 K were curated from the ILThermo database: one containing 434 systems for density and another with 293 systems for viscosity. Molecular structures were optimized using the GOAT procedure at the GFN-FF level to ensure chemically realistic geometries, and a diverse set of molecular descriptors, including electronic, topological, geometric, and thermodynamic properties, was calculated. Three support vector regression models were built: two for density (IonIL-IM-D1 and IonIL-IM-D2) and one for viscosity (IonIL-IM-V). IonIL-IM-D1 uses three simple descriptors, IonIL-IM-D2 improves accuracy with seven, and IonIL-IM-V employs nine descriptors, including DFT-based features. These models, designed to predict the mentioned properties at room temperature (298 K), are implemented as interactive applications on the atomistica.online platform, enabling property prediction without coding or retraining. The platform also includes a structure generator and searchable databases of optimized structures and descriptors. All tools and datasets are freely available for academic use via the official web site of the atomistica.online platform, supporting open science and data-driven research in molecular design.