Application of Neural Networks in the Prediction of Surface Tensions of Binary Mixtures

Abstract

In this work, an artificial neural network (ANN) has been utilized to predict the surface tension of binary mixtures at different temperatures and concentrations and at atmospheric pressure. It has been shown that a multilayer perceptron network (MLP) can be trained better than other types of ANNs, and it can therefore be used as a predictive tool to predict the thermo-physical properties. In the modeling procedure, 60% of the available experimental data has been selected as the training set; the remaining data has been used to test and validate the network. After training and testing, the artificial neural network has been used for the prediction of the surface tension of a number of other systems, for which a minimum imprecision of 1.8% has been obtained. The results obtained from the trained network have also been compared to those obtained from the Sprow and Prausnitz model [Sprow, F.B., Prausnitz, J.M., Surface tensions of simple liquid mixtures. Trans. Faraday Soc. 1966, 62a 1097–1104]. It has been shown that the trained MLP network can predict the experimental data better than the r conventional neural network method while only a minimum number of adjustable parameters have been used, compared to the number of adjustable parameters in the thermodynamics models, such as the Sprow and Prausnitz model.

Molecular Fingerprint-Based Artificial Neural Networks QSAR for Ligand Biological Activity Predictions

“…17–23 ANN have several unique attributes which make them robust for non linear generalization problems with multidimensional inputs. For instance, the networks have adaptive learning behaviors in which they learn from previous examples and adapt to changes in input parameters.…”

Surface Tension Estimation of Binary Mixtures of Organic Compounds Using Artificial Neural Networks

“…Thus, the second model with the parameters listed in Table II In order to clarify the ability of the present model, a comparison between the present model and previous works, including thermodynamic modeling and artificial neural networks (Di Nicola and Pierantozzi's model (2013), the model proposed by Parhizgar et al (2012) and Escobedo and Mansoori's model (1998)) is made for a number of mixtures at the same temperature ranges, as seen in Table IV. These comparisons are a testament to the reliability of the new proposed model.…”

“…Since both the present model and the model of Parhizgar et al (2012) are based on the ANNs, these models could be compared in more details. The present model has one hidden layer with a total number of weights and biases of 91, whilst the model of Parhizgar et al (2012) has three hidden layers with 121 parameters.…”

“…The present model has one hidden layer with a total number of weights and biases of 91, whilst the model of Parhizgar et al (2012) has three hidden layers with 121 parameters. The temperature ranges in this study and the Parhizgar et al (2012) are 170 K and 150 K, respectively.…”

In Silico Prediction of Chemicals Binding to Aromatase with Machine Learning Methods

“…45,46 The artificial NN can "learn" and "recognize" non-linear functions and recognizing patterns, ideally can "fit" them into a wide range of applications in complex systems.…”