Prediction of liquid viscosity of pure organic compounds via artificial neural networks

Neural network models have been developed to estimate liquid viscosity of p ure organic compounds at ambient temperature. These models employ different descriptors as characterizing parameters of the compounds. Three judgement criteria were imposed upon the proposed models: the accuracy of the obtain ed results, the type and number of the descriptors used as input parameters . The relative importance of the input variables was assessed. In all the c ases analyzed, easily accessible properties of the organic compounds have b een chosen as input parameters to train the neural network models. The numb er of the input properties was limited to a minimum without sacrificing the accuracy of the results. A set of 110 data points covering a wide variety of organic compounds with a viscosity range of 0.197-19.9 mPa.s was employed in training the neural n etwork models. The validity of the models was tested using 35 data points t hat were not included in the training set. The obtained results were compar ed with predictions from various published models. The neural network model s have the advantage of providing accurate results for a wide spectrum of s tructures of organic compounds using readily available physicochemical prop erties.