Application of machine learning in moisture content prediction of coffee drying process

Abstract

Humidity prediction is not only the key success factor in maintaining product quality and shelf life but also an opportunity to save energy and steam in drying technology. The purpose of this thesis is to determine the predictive effectiveness of machine learning techniques in the coffee manufacturing process. The beverage company that specialized in coffee drying techniques provided the practical data for this thesis. Although they have invested a lot in their database, it has not yet been fully utilized in the investigation of the relationship between process parameters and moisture prediction. There are two supervised machine learning techniques (ANN and ANFIS) are conducted, two transformation data techniques (Normalization and Standardization), two evaluation measures (MAE and RMSE), and two feature selection techniques (Mutual Information and Correlation). As the result, the ANFIS method shows an outweigh performance in predictive food drying techniques, especially coffee products.in