Circuit-based neural network models for estimating the solubility of diosgenin.

Several new circuit-based neural network models were conceived and utilized to estimate the solubility of diosgenin. Six mixed alcohol solvents and a pure solvent (carbon tetrachloride) were selected as the model systems to demonstrate the point of interest. To make full use of the collected solubility data of diosgenin in these solvents, they were categorized into training, testing and validation sets and a 5-fold cross validation was adopted in the buildup of the model. The results of the statistical analysis and the sum of ranking differences method indicate the parallel-serial neural network model gives more accurate description of the solubility data of diosgenin in contrast to other patterns. It also outperforms two empirical equations in terms of calculating accuracy. In addition, this suggested model can exhibit the effect of the changes of the components and their proportions in solvent on the solution behavior of diosgenin correctly. [ABSTRACT FROM AUTHOR]