Prediction of the electric conductivity of ionic liquids by two chemometrics methods

In recent years, the study of the properties of ionic liquids (ILs) and their structures has developed largely. Among the common physicochemical properties of pure ILs, electric conductivity (EC) is of crucial importance for both practical and fundamental viewpoint. In order to develop effective models for predicting the EC value of various ILs, the relationship between the structural descriptors and the EC of thirty-five ionic liquids at different temperatures was investigated by multi-linear regression (MLR) and a back propagation artificial neural network (ANN). As a result, a three layer ANN with four variables selected by the MLR model as input nodes was successfully set up. The descriptors selected by MLR were suitable and significant to be the input nodes of the ANN model in this study. Moreover, the ionic conductivities calculated by the ANN model, having a high correlation coefficient and low root mean squared error, were quantitatively in good agreement with the experimental values. The ANN model was proved to be better than the MLR model.