Figure 1. The comparison of leave-one-out cross-validation (LOO-CV) and leave-one-ion-out cross-validation (LOIO-CV) methods: orange ball with Ci stands for cations, blue ball with Ai stands for anions and Ci tightly close to the Ai stand for ILs, and the balls in different shades with fk (T ,P ) represent data points at variable temperatures and pressure.
Other validation protocols. The correlation coefficient (R 2) and mean absolute error (MAE) were used to evaluate the QSPR model quantitatively35. The definitions of these parameters are listed in Table S1 of Supporting Information (Statistical parameters.docx). Through external validation, the predicting ability of these models was fully evaluated byR 2training of the training set and R 2testing of the testing set. To avoid cations and anions in the testing set reappearing in the training set in the meantime, the dataset was divided into training set (80%) and testing set (20%) by the ion structures and the proportion of data points. Y -randomization test was used to avoid the possibility of chance correlation in modeling36.