Reliable and robust f(T,P,I)-QSPR models for ionic liquids enabled by balancing data distribution and LOIO-CV

Abstract

The thermodynamic properties at variable temperature and pressure, such as density (ρ) and viscosity (η) are necessary in chemical process design. The quantitative structure-property relationship (QSPR) is a quick and accurate method to obtain the properties from a large number of potential ionic liquids (ILs). The QSPR models for ρ and η may have “pseudo-high” robustness validated by leave-one-out cross-validation (LOO-CV) and weakened stability with the unbalanced data point distribution. A rigorous model evaluation method named the leave-one-ion-out cross-validation (LOIO-CV) was proposed to evaluate robustness of ILs QSPR models. Balancing the distribution of data points in ILs, two f(T,P,I)-QSPR models were developed with norm index (I) to predict ρ and η of ILs at variable temperature and pressure. LOIO-CV method can enhance the stability QSPR model in predicting the properties of ILs with new cations and anions, which is essential for data driven design of ILs.

QSPR Model to Predict the Speed of Sound of Ionic Liquids as a Function of Variable Temperature and Pressure

“…For internal validation, the leaveone-ion-out cross-validation (LOIO-CV) with stringent rules is adopted, rather than the leave-one-out cross-validation (LOO-CV) and K-fold cross-validation (K-CV) adopted in most previous reference studies. 25 The LOIO-CV takes ILs with the same ion (cation or anion) as the test set to verify the robustness of the model, which includes the leave-one-anionout cross-validation (LOAO-CV) and the leave-one-cation-out cross-validation (LOCO-CV). The use of LOIO-CV could avoid the "pseudo-high" accuracy brought by LOO-CV and K-CV in evaluating QSPR models for ILs.…”