Optimization methodology of artificial neural network models for predicting molecular diffusion coefficients for polar and non-polar binary gases

In this study, an artificial neural network (ANN) is used to develop predictive models for estimating molecular diffusion coefficients of various gases at multiple pressures over a large field of temperatures. Two feed-forward neural networks NN1 and NN2 are trained using six physicochemical properties: molecular weight, critical volume, critical temperature, dipole moment, temperature, and pressure for NN1 and molecular weight, critical pressure, critical temperature, dipole moment, temperature, and pressure for NN2. The diffusion coefficients are regarded as the output. A set of 1252 gases (941 non-polar and 311 polar gases) is used for training and testing the ANN performance, and good correlations are found ($R=0.986$ for NN1 and $R=0.988$ for NN2). The result of the sensitivity analysis shows the importance of the six input parameters selected for modeling the diffusion coefficient. Moreover, the present ANN model provides more accurate predictions than other models.