Intelligent Modeling of Moisture Sorption Isotherms in Indian Milk Products Using Computational Neuro-genetic Algorithm

Abstract

A hybrid Computational Neuro-genetic Modeling (CNGM) algorithm has been described for modeling moisture sorption isotherms in two industrially important Indian milk products, viz., dried acid casein powder and milk- and pearl millet-based weaning food called “fortified Nutrimix” powder. Casein isotherms were studied at three temperatures, i.e., 25, 35, and 45 °C. Nutrimix isotherms were considered at four temperatures, i.e., 15, 25, 35, and 45 °C. Isotherms of aforementioned products were measured over water activity range of 0.11–0.97. The neuro-genetic models were developed using a novel algorithm, which was utilized for training neural networks rather than traditional learning algorithms like error back-propagation technique. Also, conventional two-parameter empirical models, viz., Brunauer–Emmett–Teller (BET), Caurie, Halsey, Oswin, and Smith; and/or three-parameter models, viz., modified Mizrahi and Guggenheim–Anderson–de Boer (GAB) models were considered from elsewhere (that were fitted to same data as used in this study) for comparison of neuro-genetic models’ prediction potential. Accordingly, neuro-genetic and GAB (best among conventional models considered) models predicted sorption isotherms with accuracy, in terms of root-mean-squared percent error, ranging as 0.17–0.26 and 1.93–5.78 for adsorption, and 0.17–0.39 and 1.40–5.01 for desorption, respectively, in case of casein; and 0.04–0.17 and 5.48–10.60 for adsorption, and 0.06–0.15 and 5.54–9.54 for desorption, respectively, for Nutrimix. Evidently, neuro-genetic models outperformed conventional empirical sorption models. Hence, it is deduced that hybrid CNGM approach is potentially intelligent precision modeling tool for predicting adsorption and desorption isotherms in Indian milk products, i.e., dried acid casein powder and “fortified Nutrimix” powder.