Solubility prediction of anthracene in binary and ternary solvents by artificial neural networks (ANNs)

Abstract

Solubility of anthracene in binary and ternary solvent systems was modeled by artificial neural networks (ANNs) technique. The results obtained using the ANN method indicated that the solubility of anthracene in mixed solvents could be calculated using the mole fraction solubilities in pure solvents, mole/volume fraction of solvents and solvent's solubility parameters. The topology of neural network was optimized empirically and optimum topology was a 6-6-1 architecture for binary and 9-6-1 for ternary mixtures. The solubility of anthracene in mixed solvents was estimated by means of ANN and the predicted solubility was compared with experimental solubility data. The overall absolute percentage mean deviation (OPMD) for trained ANNs using all data points in 25 binary and 30 ternary solvent systems were 0.16 and 0.20%, respectively. A minimum number of data points from binary and ternary solvents have been employed to train the ANN and solubility at other solvent compositions has been predicted. The OPMD obtained for solubility in binary and ternary solvents were 0.67 and 0.27%, respectively. The trained network with 25 binary data sets was applied to predict the solubility in 16 other binary solvent systems and the OPMD obtained is 15.32%. The results of ANNs were also compared with similar numerical analyses carried out using multiple linear regression models and found that the ANN method is generally promising more accurate calculations.