artificial neural network (ann) approach for modeling experimental data of viscosity and density of a ternary solution

The artificial neural network (ann) approach was used to model experimental viscosity and density data for ternary aqueous solutions of calcium chloride and potassium chloride. then, the ann model was compared with the previously investigated models e.g., modified jones-dole, hu, exponential and gf models used for the same dataset. in the present study, the levenberg-marquardt algorithm or trainlm command was selected as the training algorithm. subsequently, different configurations of the network were compared and the optimal multi-layer perceptron (mlp) network was designed with 3 hidden layers and [8 4 3] neurons since it showed better performance. moreover, 80% of the dataset for network training, 10% for validation and the rest for network testing were randomly selected. amid investigated models, ann obtained the minimum mean square error (mse) of 6.2008×10-5 and maximum r2 of 0.9997 while at best modifies jones-dole could achieve a mse of 2.768×10-5 and r2 of 0.9996. suggesting that the ann model is the most optimal model for modeling the viscosity and density of this ternary solution.