Evaluating prediction ability of different models used for solute solubility calculation in binary solvent systems

Abstract

The available theoretical and empirical models for expression a solute solubility in a binary solvent system as a function of the solvent concentration have been compared by the accuracy and predictability points of view. In order to evaluate the accuracy of the equations with different number of curve fitting parameters, 3 to 7 parameter models from the considered methods were employed using whole data points in each set. In addition, a minimum number of 3 and/or 5 data point were used to calculate the model parameters from which the whole data points in each set was predicted. Using the whole data in a set, it became evident that no single model was superior in all aspects of accuracy. However, one of the models in which the number of data expressed as percentage of all data points in all sets was adhered best whereas another model was the best when the percent overall average error for all the set was used as a measure of accuracy. When a minimum number of data in a set was employed to estimate model constants for evaluating the predictability, it became apparent that those produced accurate results when the whole data points did not yield satisfactory predictions and other models were superior in this respect.