Analytical Study of Methanol Content in Traditional Fermented Beverages: Improving Evaluation Procedures and Developing a Predictive Model using Nonlinear Regression model and Artificial Neural Networks

Abstract

Artificial intelligence has been used in many fields, as its use has led to better results than the results of traditional methods, and among the fields in which it can be used is the field of food science. In this research, methanol levels in fermented beverages were predicted in order to evaluate their safety by proposing two prediction models: a Nonlinear Regression Model and an Artificial Neural Network Model .

In order to obtain the input variables, the methanol concentration was measured for 32 fermented samples, which took 15 to 30 days of fermentation, in media with temperatures of 10-30 °C. Thus, the input variables are the type of fermenting liquid, PH and fermentation temperature, fermentation time, and the sterilization process of the fermentation medium. The concentration of methanol formed during fermentation was used as a dependent variable for the nonlinear regression model and as an output for the neural network model. The input data was processed and the neural network model outperformed the nonlinear regression model in estimating and predicting methanol levels because it had a greater R² value and lower MSE, RMSE, and MAE values compared to the nonlinear regression model.