Statistical modeling of aspirin solubility in organic solvents by Response Surface Methodology and Artificial Neural Networks

Abstract

The present work is aiming at statistical modeling and prediction of solubility of aspirin based on two intelligent methods including Response Surface Methodology (RSM) and Artificial Neural Networks (ANN). To develop the models, a data bank including 109 data belonging to the solubility of aspirin in ethanol, acetone, 2-propanol, 1-octanol, ethyl acetate, isobutanol, isobutyl acetate, 1-butanol, MIBK and propylene glycol as organic solvents was extracted from the literature. Temperature, molecular weight of the solvents, critical pressure and temperature and acentric factor were chosen as independent variables for the modeling. Both RSM and ANN models were statistically compared using coefficient of determination (R2), Root Mean Square Error (RMSE), Average Absolute Deviation (AAD%) and Sum of Absolute Residual (SAR) obtained for the data set. R2 and A.A.D% were determined as 0.9992 and 2.598% for ANN, and 0.997 and 3.884% for RSM model, respectively. It was identified that both developed model can accurately predict the solubility of aspirin in different organic solvents, however, ANN was more accurate due to its topology and structure, which promotes the accuracy of the model. The correlation was also verified with seven more experiments. It was found that the proposed statistical RSM model is able to obtain the solubility of aspirin in various organic solvents using extrapolation and/or interpolation feature.