Extensive Databases and Group Contribution QSPRs of Ionic Liquids Properties. 2. Viscosity

New quantitative
structure–property relationships (QSPRs)
for estimating dynamic viscosity (η) of pure ionic liquids (ILs)
as a function of temperature and group contributions (GCs) are presented
and evaluated. The correlations were established using three common
machine learning algorithms (stepwise multiple linear regression,
feed-forward artificial neural network, and least-squares support
vector machine) on the basis of the largest database reported thus
far, including the data for 2068 distinct ILs (3236 data sets and
22 268 data points). The GC scheme as well as two-stage modeling
protocol (representing the property using separate reference term
and temperature correction models) were applied consistently with
the previous contribution [Ind. Eng. Chem. Res. 2019, 58, 5322–5338]. Standard internal
and external validation techniques (such as, K-fold
cross-validation, y-scrambling, “hold-out”
testing, and the Williams plot) were adopted to select the best set
of GCs, hence statistically the most significant model. The impact
of the chemical structure of both cations and anions (as well as their
combination) on the accuracy of prediction and classification (with
respect to the order of magnitude of η) is analyzed in detail.
The obtained models are compared with other methods reported in the
literature. In particular, a broad comparison of the finally recommended
model with the QSPR, employing descriptors derived from molecular
geometry and charge distribution [J. Phys. Chem. B 2011, 115, 300–309] is given.