Prediction of ionic conductivity of imidazolium-based ionic liquids at different temperatures using multiple linear regression and support vector machine algorithms

Abstract

Ionic liquids (ILs) are well recognized as promising and environmentally friendly solvents owing to their remarkable features, which have captured the imagination of the global research community to expand ILs’ usage across various industrial processes and applications. Nevertheless, the formulation of ILs with desired properties requires strenuous effort in discovering the suitable combination of cations and anions. Hence, it is imperative to develop simple and accurate models to predict the physicochemical properties of ILs prior to experimentation. Ionic conductivity is one of the most significant intrinsic properties which affects the transport capabilities of ILs and it has been a limiting factor in the design of suitable ILs. In the present study, the conductivity of different imidazolium-based ILs has been estimated and correlated via the Quantitative Structure–Property Relationship (QSPR) approach using two different algorithms, namely multiple linear regression (MLR) and support vector machine (SVM) regression coupled with stepwise model-building. A set of descriptors, including interaction energies as well as dielectric energy of the ILs’ cation–anion pairs generated by the Conductor-like Screening Model for Real Solvents (COSMO-RS), were employed to derive the best-fit model. The models were developed using experimental data of imidazolium-based ionic liquids collected from the literature with conductivity in the range of 0.008–5.1 S m⁻¹ at temperatures between 268.15 K and 398.15 K. The coefficients of determination (R²) for the MLR and SVM model's entire data set are 0.8556 and 0.9906, respectively, while the average absolute relative deviations (AARD) are 46.55% and 7.15%, respectively. This suggests that the non-linear model developed using the SVM regression algorithm fits better with the conductivity data set and is more reliable than the MLR algorithm. The stepwise approach reveals that conductivity is highly affected by van der Waals forces and temperature, followed by electrostatic forces and dielectric energy to some extent. The prediction results from this work will aid the screening process of suitable ILs with desired conductivity for specific applications.