Automated Prediction of Brewing Methods for Coffee Flavors by Multiple Linear Regression, Support Vector Regression, Random Forest, and Neural Network

Many studies have been done to identify the correlation between volatile components in brewed coffee liquid and different coffee brewing parameters and methods, but few are focusing on taste and non-volatile components. Our work selected 15 non-volatile taste-determining chemical components in brewed Arabica coffee under different brewing parameters (water temperature and water/coffee ratio) and quantitatively measured the components using GCMS. The taste activity values (TAVs) of 15 chemicals are calculated, among which 5 chemicals with TAV> 1 are used to train machine learning (ML) models. Among these models, Neural Network (NN) and Random Forest perform the best. And NN is used to establish the relationship between TAV of flavor-determining chemical components and the flavor of brewed coffee.