Prediction of surface tension of ionic liquid based on imidazolium using artificial neural network

Nowadays, with the progresses in technology to solve problems where there is no exact mathematical relationship between input and output, neural networks are efficiently proposed and used. In the shadow of its unique features, in this study, two multilayer perceptron neural networks including feedforward artificial neural network (FFANN) and cascade artificial neural network (CANN) were proposed to predict the surface tension of imidazolium-based ionic liquids. To verify the validity of the proposed models, 1251 experimental data points were collected from various previously published literature including the surface tension of 40 ionic liquids in a wide range of temperatures (from 263.61 to 533.2 K). The results showed that the proposed CANN consists of three inputs including molecular weights of anionic and cationic part of ionic liquid and temperature with a hidden layer containing 8 neurons with a hyperbolic tangent activation function and trained with Levenberg–Marquardt algorithm has the best correlative capability for surface tension of ionic liquids. In addition, error analysis of test data set with an average absolute relative deviation percent of 1.07 indicates the appropriate performance of the nonlinear CANN model in the linking between network inputs and surface tensions. Also, comparing the accuracy of the proposed model with existing models, including the corresponding states principle, Parachor, the group method of data handling (GMDH) and the model based on least-squared supported vector machine (LSSVM) indicate the superiority of the proposed model.