Non-linear models applied to experimental spectroscopical quantitative analysis of aqueous ternary mixtures of imidazolium and pyridinium-based ionic liquids

The study of a ternary system formed by two ionic liquids (ILs) (N-butyl-pyridinium chloride ([bpy][Cl]) and 3-butyl-1-methylimidazolium chloride ([bmim][Cl])) and water has been carried out using UV–vis absorption measurements. The results were employed to design non-linear mathematical models based on artificial neural networks. The models provided accurate results for the estimation of the concentration of [bpy][Cl] and [bmim][Cl] (the best model offered the following mean prediction errors (MPEs): MPE[bpy][Cl] = 3.81% and MPE[bmim][Cl] = 5.78%). Different methodologies have been considered to describe the system, which has led to discussions regarding the relevance of number, type, and quality of independent variables used in the models. This study allowed improving the estimations achieved by different models and decreasing the error attained seven- and twelve-fold for the concentration estimation of [bpy][Cl] and [bmim][Cl], respectively, from the worst to the best model.