Optimization of Millet Malting Parameters Using Artificial Neural Network and Response Surface Methodology

Abstract

The quality of malt produced from cereals is significantly influenced by various factors, including steeping and germination periods. Monitoring these factors and their effects on malt grain characteristics is often time-consuming and costly. In this context, this study aimed to predict trends in changes to certain characteristics of millet-derived malt, influenced by varying steeping durations (24–48 h) and germination times (5–9 days). Changes in these characteristics were predicted using response surface methodology (RSM), which incorporated a central composite design and an artificial neural network (ANN). The findings indicated that increasing the steeping and germination durations led to a decrease in malting efficiency, thousand grain weight, and true density of the samples. Conversely, the cold-water extract efficiency, the Kolbach index, and the extract color increased. The optimization process revealed that to achieve the highest-quality malt, the steeping duration should be 42.54 h, followed by a germination period of 5 days. Under these conditions, the malting efficiency reached 75.44%, with a thousand grain weight of 4.85 g, a true density of 977.43 kg/m³, a cold-water extract efficiency of 9.19%, a Kolbach index of 32.45%, and an extract color value of 13.87. An analysis of different neural networks revealed that the feed-forward backpropagation network with a 2-6-6 topology was the best-performing model. This network achieved a correlation coefficient greater than 0.999 and a mean squared error of less than 0.00001. It employed the hyperbolic tangent sigmoid transfer function, the resilient backpropagation learning algorithm, and 1000 learning cycles. Furthermore, a comparison of the correlation coefficients derived from the RSM and the ANN demonstrated that the ANN method is superior for predicting changing trends in millet grains during the malting process.